Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
COOPERATIVE LEARNING NEURAL NETWORKS AND SYSTEMS
Document Type and Number:
WIPO Patent Application WO/2019/046103
Kind Code:
A1
Abstract:
Systems, methods, and apparatuses related to cooperative learning neural networks are described. Cooperative learning neural networks may include neural networks which utilize sensor data received wirelessly from at least one other wireless communication device to train the neural network. For example, cooperative learning neural networks described herein may be used to develop weights which are associated with objects or conditions at one device and which may be transmitted to a second device, where they may be used to train the second device to react to such objects or conditions. The disclosed features may be used in various contexts, including machine-type communication, machine-to-machine communication, device-to-device communication, and the like. The disclosed techniques may be employed in a wireless (e.g., cellular) communication system, which may operate according to various standardized protocols.

Inventors:
LUO, Fa-Long (1144 Little John Way, San Jose, California, 95129, US)
SCHMITZ, Tamara (195 Tucker Road, Scotts Valley, California, 95066, US)
CHRITZ, Jeremy (762 Hayes Street, Apt. 36Seattle, Washington, 98109, US)
CUMMINS, Jamie (782 Cherry Avenue, Bainbridge Island, Washington, 98110, US)
Application Number:
US2018/047783
Publication Date:
March 07, 2019
Filing Date:
August 23, 2018
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
MICRON TECHNOLOGY, INC. (8000 South Federal Way, Boise, Idaho, 83716-9632, US)
International Classes:
H04W4/46; G06N3/08; H04W4/02; H04W4/38
Foreign References:
US9147353B12015-09-29
US20050060069A12005-03-17
US5377108A1994-12-27
US20100191391A12010-07-29
US20140056229A12014-02-27
Attorney, Agent or Firm:
FAUTH, D. Justen et al. (Dorsey & Whitney LLP, IP Docket - SE701 5th Avenue, Suite 610, Seattle Washington, 98104, US)
Download PDF:
Claims:
What is claimed is:

1 . A method comprising:

identifying, at a first wifeless communication device, first data from a sensor of a first wireless communication device;

receiving, at the first wireless communication device, second data transmitted from a second wireless communication device; and

training a cooperative learning neural network at the first wireless communication device using a combination of the first data and the second data.

2. The method of claim 1, wherein the first and second wireless communication devices comprise vehicles.

3. The method of claim 1 , wherein the first data or tlie second data, or both, comprise image data.

4. The method of claim 1 , wherein training trie cooperative learning neural network living the combination of the first data and the second data occurs when the second wireless communication, device is within a threshold distance of the first wireless communication device.

5. The method of claim 4, further comprising determining that the second wireless communication device is within the threshold distance.

6. The method of claim 1, wherein said training comprises generating weights for the cooperative learning neural network.

7. The method of claim 1 , wherein the weights are indicative of an object.

8. The method of claim 1, further comprising using the cooperative learning neural network to identify at least one crash condition.

9. The method of claim 8. further comprising adjusting a speed or direction, or both, of the first wireless communication device responsive to identifying the at least one crash condition.

10. The method of claim 1 , wherein said training comprises unsupervised training based on operation of the cooperative learning neural network.

1 1. The method of claim 1 , wherein said (raining comprises supervised training based on a set of references.

12. The method of claim 1 , wherein receiving the second data comprises: receiving the second data from a node of a wireless communication network configured according to at least one of a 3GPP New Radio (NR) specification, a 3GPP Long Term Evolution (LT E) specification, a MulteFire specification, a IEEE 802.1 1 specification, or any combination thereof.

13. A wireless communication device comprising:

a sensor;

a transceiver, the transceiver configured to transmit first vehicle data from the sensor and receive second vehicle data from at least one other sensor; and

a cooperative learning neural network, the cooperative learning neural network configured to utilize both the first vehicle data and the second vehicle data to train the cooperative learning neural network.

14. The wireless communication device of claim 13, wherein the cooperative learning neural network is configured to calculate weights of the cooperative Learning neural network based at least on the second vehicle data.

15. The wireless communication device of claim 14, wherein the transceiver is further configured to receive third vehicle data from another sensor,, and wherein the cooperative learning neural network is configured to calculate the weights based further on the third vehicle data.

16. The wireless communication device of claim 13, further comprising a controller coupled to the cooperative learning neural network, and wherein the controller i,s configured to adjust a direction or speed, or both, of the vehicle responsive to the cooperative learning neural network,

17. The wireless communication device of claim 13, wherein the cooperative learning neural network is configured to identify a precursor to a crush condition ,

18. The wireless communication device of claim 17, wherein the crash condition comprises at least one of a rear-end collision, a collision with a parked vehicle, a side-impact crash, an animal collision, or hydroplaning, or any combination thereof.

1 9. A method comprising:

receiving raw sensor data from a first wireless communication device at a second wireless communication device;

utilizing the raw sensor data at the second wireless communication device to train a cooperative neural network; and

utilizing the cooperative neural network at the second wireless communication device to detect a crash condition in within u threshold time that is based at least in part on the crash condition.

20. The method of claim 19, wherein the first wireless communication device and the second wireless communication device comprise vehicles.

21. The method of claim 19, wherein the raw sensor data comprises sensor signals from one or more sensors of the first wireless communication device.

Description:
COOPERATIVE LEARNING NEURAL N ETWORKS AND SYSTEMS

[001 ] Examples described herein relate to communications capabilities of vehicles.

Examples of vehicles implementing cooperative learning neural networks are described.

BACKGROUND

[002] Machine learning for vehicle systems has utilized information sensed by the vehicle in order to make future decisions regarding that vehicle. The amount of information which may be obtained from other vehicles has been limited due to the bandwidth available in conventional communication systems. However, (his will not be the case with the deployment of 5G wireless communication infrastructure which can provide the sufficient bandwidth such as at the amount of 10 G bits per second.

[003] An example method according to an embodiment of the disclosure may include identifying, at a first wireless communication device, first data from a sensor of a first wireless communication device; receiving, at the first wireless communication device, .second data transmitted from a second wireless communication device; and training a cooperative learning neural network at the first wireless communication device using a combination of the first data and the second data.

[004 ] Additionally or alternatively, the first and second wireless communication devices include vehicles.

[ 005 ] Additionally or alternatively, the first data or the second data, or both, include image data.

[006] Additionally or alternatively, training the cooperative learning neural network using the combination of the first data and the second data occurs when the second wireless communication device is within a threshold distance of the first wireless communication device.

[007] Additionally or alternatively, the method further includes determining that the second wireless communication device is within the threshold distance. [008[ Additionally or alternatively, the training includes generating weights fur the cooperative Learning neural network.

[009] Additionally or alternatively, the weights are Indicative of an object.

[010] Additionally or alternatively, the method further includes using the cooperative learning neural network to identify at least one crash condition.

[011 ] Additionally or alternatively, the method further includes adjusting u speed or direction, or both, of the first wireless communication device responsive to identifying the at least one crash condition.

[012] Additionally or alternatively, the training includes unsupervised training based on operation of the cooperative learning neural network.

[013] Additionally or alternatively, the training includes supervised training based on a set of references,.

[014] Additionally or alternatively, receiving the second data includes receiving the second data from a node of a wireless communication network configured according to at least one of a 3GPP New Radio (NR) specification, a 3GPP Long Term Evolution (LTE) specification, a MultcFire specification, a IEEE 802.1 1 specification, or any combination thereof.

[015] An example wireless communication device according to an embodiment of the disclosure may include a sensor; u transceiver; and a cooperative learning neural network. The transceiver is configured to transmit first vehicle data from the sensor and receive second vehicle data from at least one other sensor. The cooperative learning neural network utilizes both the first vehicle data and the second vehicle data to train the cooperative learning neural network.

[016] Additionally or alternatively, the cooperative learning neural network is configured to calculate weights of the cooperative learning neural network based at least on the second vehicle data.

[017] Additionally or alternatively, the transceiver is further configured to receive third vehicle data from another sensor, and wherein the cooperative learning neural network is configured to calculate the weights based further on the third vehicle data,

[018] Additionally or alternatively, the wireless communication device includes a controller coupled to the cooperative learning neural net-work, and wherein the controller is configured to adjust a direction or speed, or both, of the vehicle responsive to the cooperative Learning neural network.

[019] Additionally or alternatively, the cooperative learning neural network is configured to identify a precursor to a crash condition,

[020 ] Additionally or alternatively, the crash condition includes at least one of a rear- end collision, a collision with u parked vehicle, a side-impact crush, an animal collision, or hydroplaning, or any combination thereof.

[021] An example method according to an embodiment of the disclosure may include receiving raw sensor data from a first wireless communication device al a second wireless communication device; utilizing the raw sensor data a: the second wireless communication device to train a cooperative neural network; and utilizing the cooperative neural network at (he second wireless communication device lo detect a crash condition in within a threshold time that is based at least in part on the crash condition.

[022] Additionally or alternatively, the first wireless communication device and the second wireless communication device include vehicles

[023] Additionally or alternatively, the raw sensor data includes sensor signals from one or more sensors of the first wireless communication device.

BRIEF DESCRIPTION OF THE DRAWINGS

[024] fiG. 1 is a schematic illustration of a system arranged in accordance with examples described herein.

[025 ] FIG, 2 is a flowchart illustrating a method arranged in accordance with examples described herein.

[026] FIG. 3 is a schematic illustration of an example system arranged in accordance with examples described herein,

[027] FIG. 4 is a schematic illustration of a cooperative learning neural network arranged in accordance with examples described herein.

[028] FIG. 5 is a schematic illustration of a wireless communications system arranged in accordance with aspects of the present disclosure. [029 ] Cooperative learning neural networks may employ machine learning techniques to make decisions based on information gathered from multiple sources that communicate with one another wirelessly. The systems described herein may leverage advanced wireless communication protocols to communicate directly between devices, in machine-to-machinc (M2M) fashion, or via a wireless communication network using machine-type communication (MTC), Such advanced wireless communication protocols may allow for high bandwidth or low latency communication that will allow physically separate or remote sensors to gather information and train a neural network.

[030] For example, cooperative learning neural networks described herein may be used to develop weights which are associated with objects or conditions at one device and which may be transmitted to a second device, where they may be used to train the second device to react to such objects or conditions. This is in contrast to prior machine learning solutions in which physically separated sensors could, not effectively or timely communicate with one another to allow for training a neural network. The techniques described herein may be applied in n vehicle-to-vehicJe context, or they may be employed in other M2M or device-to-device (D2D) schemes.

[031 ] Figure 1 is a schematic i (lustration of a system arranged in accordance with examples described herein. The system 100 includes vehicle 102, vehicle 104, and other computing sy stents) 1 16. The vehicle 102 may include sensors) 108, transceiver 1 10, controller 132, and cooperative learning neural network 126, or some combination of such devices. Some or all of sensor(s) 108, transceiver 1 10, controller 132, and cooperative learning neural network 126 may be components of a .subsystem of vehicle 102. The cooperative learning neural network 126 may he implemented using processing untt(s) 112 and memory 1 14, which may store weights 106, The vehicle 104 may include sensor{s) 1 18, transceiver 120, controller 134, and cooperative learning neural network 128, or some combination of such devices, Some or all of sensors) 1 18, transceiver 120, controller 134, and cooperative learning neural network 128 may be components of a subsystem of vehicle 104. The cooperative Learning neural network 128 may be implemented using processing unit(s) 122 and memory 124, which may store weights 130. While described as vehicles in Figure 1 , vehicles 102 and/or 104 may be generally implemented by one or more wireless communication devices. [032 ] While two vehicles ore shown in Figure 1 , any number may be present in other examples including 3, 4, 5, 6, 7, 8, 9, 10, 50, 100, 500, 1000, or another number of vehicles, Any of a variety of vehicle types may implement systems and methods described herein including, but not limited to, automobiles (e.g., passenger cars, taxis, ride share vehicle.), trucks (e.g., semi-trucks, heavy equipment, service trucks such as but not limited to delivery trucks, garbage trucks, postal vehicles), busses, trains, driverless vehicles * aircraft, helicopters, ships, dirigibles, space vehicles, rockets, submarines, or combinations thereof.

[033] Vehicles described herein, such as vehicle 102 and. vehicle 104 of Figure 1 , may be equipped with one or more sensors, such as sensor(s) 108 and sensor(s) 1 18 of Figure 1 , The sensors may be integrated with the vehicle and placed on, in, around, above, below, und/or proximate to the vehicle. Each sensor may be used to sense one or more environmental parameters. Examples of sensors which may be used include, but are not limited to, optical sensors (e.g., cameras, infrared sensors), temperature sensors, acceleromcters, gyroscopes, inertia! sensors, humidity sensors, RADAR sensors (e.g., LiDARs), ultrasonic sensors, vehicle parameter sensors (e.g. , sensors used to sense a position and/or state of one or more vehicle components, such as but not limited to, a wheel position sensor, a wheel speed sensor, an airspeed sensor), altimeter, or combinations thereof,

[034 ] Vehicles described herein, such «s vehicle 102 and vehicle 104 of Figure 1 , may be equipped with one or more transceivers. The transceiver may include one or more receivers, transmitters, or combinations thereof. Transceivers described herein may be used to transmit and/or receive data from a variety of sources, including other vehicles and/or other computing systems. For example, the transceiver 1 10 may be used to transmit vehicle data pertaining to vehicle 102, which data may be generated by sensor(s) 108, The transceiver 1 10 may be used to receive vehicle data pertaining to vehicle 104 or other vehicles, which may be generated by sensor(s) 1 18, Data transmitted by the transceiver 1 10 may be received by vehicle 104 and/or other computing systcm{s) 1 16. Transceiver 110 may be used to receive data from other computing system(s) 1 16 in some examples, In an analogous manner, the transceiver 120 may be used to transmit vehicle data pertaining to vehicle 104, which may be generated by sensor(s) 1 18. The transceiver 120 may be used to receive vehicle data pertaining to vehicle 102, which may be generated by sensor(s) .108. Data transmitted by the transceiver 120 may be received by vehicle 102 and/or other computing system(s) 1 16. Transceiver 120 may he used to receive data from other computing system(s) 1 16 in some examples.

[035 ] Transceivers described herein generally transmit and/or receive data using wireless communication techniques, f.n some examples, transceivers, such as transceiver 110 and transceiver 120 may communicate using 5G wireless communication techniques. 50 wireless communication techniques may, in some examples, allow for adequate bandwidth and speed of communications such that sensor data from one vehicle (e.g., vehicle 104) may timely be received by another vehicle (e.g., vehicle 102) and utilized by the receiving vehicle to impact the operation of the receiving vehicle (e.g., driving). Jn some examples, transceivers described herein, such as transceiver 1 10 and/or transceiver 1 20 may utilize full duplex communication techniques, including full duplex SG wireless communication techniques. Accordingly, the transceiver 1 10 and/or transceiver 120 may substantially simultaneously both transmit and receive data in sonic examples. The transceiver 1 10 and/or transceiver 120 may be equipped with interference cancellation circuitry which may facilitate the simultaneous transmission and receipt of data in some examples.

[036| Vehicles described herein, such as vehicle 102 and vehicle 104 of Figure 1 , may include one or more controllers, such as controller 132 and controller 134 shown in Figure 1 . The controllers may provide control signals to one or more components of the vehicle, including control signals used to control operation of the vehicle. For example, the controllers may provide control signals to set and/or adjust a speed and/or heading of the vehicle. Other aspects of the vehicle may be controlled by controllers described herein (e.g., the speed, setting, heading, or operation of any component of the vehicles). Accordingly, the controllers, such as controller 132 and/ov controller 134 may be in electrical and/or mechanical (e.g., pneumatic) communication with a variety of components of the vehicle. In some examples, multiple controllers may be provided in each vehicle. Controllers described herein may be in communication, either directly or indirectly through other vehicle components, with sensors of the vehicle, such us sensors) 108 and/or sensor(s) 1 18. In this manner, readings from the sensors may be used as inputs to the controller 134 which may provide control signals according to sensor readings. Controllers described herein may be implemented using any number or variety of processing unit(s), including but not limited to, processors, circuitry (e.g., application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs)), nnd'or .microcontrollers.

[037] Examples of vehicles described herein may implement one or inore cooperative learning neural networks, such as cooperative learning neural network 126 and cooperative Learning neural network 128 of Figure 1. Cooperative learning neural networks described herein generally utilize a combination of input from the vehicle in which they arc operating and input from one or more other vehicles to implement and/or train the cooperative learning neural network. For example, cooperative learning neural networks described herein may utilize a set of weight or weights to process inputs and provide an output.

[038] Within a cooperative learning neural network, a number of sets of weights may he used to process an input, with each set of weights being indicative of a particular condition to be identified in the input. For example, a set of weights may be associated with each object in an image that it may be desirable for the cooperative learning neural network to identify. A set of weights in different values may be associated, for example, with each of a stop sign, a street sign, a vehicle, a person, a building, a traffic Light, a bird, or an animal.

[039] Sets of weights may be provided to or associated with other objects in other examples. Any number of sets of weights may be used to identify objects in images as described herein, in some examples, a set of weights may be associated with i\ condition that it may be desirable for the cooperative learning neural network to identify. A set of weights in different values may be associated, for example, with a crush condition, such as but not limited to, a rear-end collision, a side-impact crash, an animal condition, a collision with a parked vehicle, and/or a hydroplaning condition, The cooperative learning neural network may be processed with each of n number of sets of weights and a condition and/or object may be recognized by the cooperative learning neural network according to the set of weights which produces a particular output (e.g., a maximum output, a minimum output, tind/or a predetermined output).

[040 ] Cooperative learning neural networks may be implemented using one or more processing units and memory. For example, the cooperative Learning neural network 126 may be implemented using processing unit(s) 112 and memory 114, which may store weights 106. The cooperative learning neural network 128 may be implemented using processing unit(s) 122 and memory 124, which may store weights 130. Processing unite which may be used lo implement cooperative learning neural networks described herein may be implemented using any number or variety of processing unity(s), including but not limited to, processors, circuitry (e.g., application specific integrated circuits (ASICs), field programmable gale arrays (FPGAs)), and/or microcontrollers. Memory which may be used to implement cooperative learning neural networks described herein generally may include any number or type of memory devices, including but not limited to, RAM , ROM, DRAM, Flash, or other memory types or devices. Memory used in cooperative Learning neural networks may store weights which may be used to process inputs and provide outputs from cooperative teaming neural networks.

[041 ] Cooperative learning neural networks described herein may utilize input from both the vehicle in which they are operating and another vehicle (aiid or multiple other vehicles) in training the cooperative learning neural network and/or in operating the cooperative learning neural network, In some examples, cooperative learning neural networks described herein may be utilized to identify objects in images. During operation, image data may be received by cooperative learning neural network 126 from the vehicle 102, e.g., from sensor(s) 108. Image data may also be received by cooperative learning neural network 126 from the vehicle 104, e.g. , from sensor(s) 1 18 through transceiver 120 and transceiver 110, Both the image data received from the vehicle 102 and the image data received from the vehicle 104 may include image data pertaining to a particular object (e.g., a stop sign). For example, the image data received from the vehicle 102 may include image data relating to a stop sign from one vantage point, while the image data received from the vehicle 104 may include image data relating to the same slop sign from a different vantage point. Both sets of i mage data, separately or in combination, may be used to train the cooperative learning neural network 126 to develop a set of weights indicative of the object (e.g., a slop sign), In this manner, the weights developed may allow the cooperative learning neural network 126 to more accurately identify a stop sign in image data than if only image data from vehicle 102 had been used,

[042| In some examples, cooperative learning neural networks may utilize only input data from vehicles which are determined to be relevant to an inquiry of the cooperative learning neural network. Accordingly, cooperative learning neural networks may evaluate an identity and/or location of a vehicle providing sensor data and may determine whether to use that sensor dala based on the identity and/or location (or other parameter). For example, consider Ihe cooperative learning neural network 126 being used to ident ify objects in image dare. The cooperative learning neural network 126 may utilize image data transmitted only from vehicles which are sufficiently nearly that they are likely to have relevant image data (e.g., they are likely to huve captured image dala relating to a same object), For example, the cooperative learning neura! network 126 may receive a location of vehicle 104 and may only utilize image data received from vehicle 104 when a distance between vehicle 102 and vehicle T04 is less than a threshold distance.

[043 ] lu some examples, cooperative learning neural networks described herein may be utilized to identify a condition (e.g., a crash condition and/or a precursor to a crash condition), Examples of precursors to crash conditions include conditions which are likely to occur prior to a crash condition. Some examples of crash conditions which muy commonly cause vehicle crashes include, but are not limited to, rear-end collision, side- impact crash, animal collision, parked vehicle collision, and hydroplaning. Sensor data muy be received by cooperative learning neural network 126, for example from sensor(s) 108, which is relevant to the crash condition. For example, considering a rear-end collision, sensor data may be provided to cooperative learning neural network 126 which includes image data from a rear-view camera on vehicle 102, Sensor data from vehicle 104 may be provided to cooperative learning neural network 126 (e.g., from sensor(s) 118 through transceiver 120 and transceiver 1 10), for example, from a front-facing camera on the vehicle 104 when the vehicle 104 is behind the vehicle 102. Additional sensor data provided from vehicle 104 may include vehicle speed. Both sets of sensor data, separately or in combination, may be used to train the cooperative learning neural network 126 to develop a set of weights indicative of a precursor to the collision condition (e.g., a precursor to a rear-end collision), in this manner, the weights developed may allow the cooperative learning neural network 126 to more accurately identify a precursor to a collision condition from sensor data than if only sensor data from vehicle .102 had been used.

[044 ] In some examples, cooperative learning neural networks may utilize input data from vehicles which are determined to be relevant to an Inquiry of the cooperative learning neural network. Accordingly, cooperative learning neural networks may evaluate an identity and/or location of a vehicle providing sensor data and may determine whether to use that sensor dala based on the identity and/or location (or other parameter ). For example, consider Ihe cooperative learning neural network 126 being used to ident ify a rear-end collision, The cooperative learning neural network 126 may utilize image dala transmitted only from vehicles which are sufficiently positioned that they arc likely to huve relevant sensor data (e.g., (hey arc likely to he positioned behind the vehicle 102 and contribute to a rear-end collision). For example, the cooperative learning neural network 126 may receive a location of vehicle 104 and may utilize image data received from vehicle 104 when the vehicle 104 is less than a threshold distance behind the vehicle 102.

[045 ] In some examples, cooperative learning neural networks may receive weights from another vehicle for use tit another vehicle. For example, vehicle 104 may transmit some or all of weights 130 to vehicle 102. The vehicle 102 may store weights received from vehicle 104 in memory 1 14. Cooperative learning neural network 126 may utilize weights 106, which may include weights received from other vehicles, such as vehicle 104, (o analyze inputs.

[046] Examples of systems described herein may include one or more other computing systems, such as other computing system(s) 116 in Figure 1. The other computing system(s) 1 16 may be in communication with one or more of the vehicles described herein und may provide all or portions of the processing described herein, and/or may provide additional sensor data for use by cooperative learning neural networks described herein, Examples of other computing system(s) 1 16 which may be used include, but are not limited to, sensor devices which may transmit data from another location in the environment proximate one or more of (he vehicles (e.g., sensors along a road, at a toll booth, in pavement). Other examples of other computing system(s) 1 16 include computing resources located remotely from one or more vehicles but in electronic communication with the vehicle (e.g., one or more computing resources accessible over a network, from a cloud computing provider, or located in the environment of one or more vehicles).

[ 047 ] When a cooperative learning neural network described herein has identified a condition (e.g., identified an object in image data and/or a crash condition), controllers described herein (e.g., the controller 132 and/or controller 134 of Figure 1) may be used to control components of one or more vehicles responsive ro ihe identification, For example, the controller 132 may provide control signals to slow and/or stop vehicle 102 responsive to cooperative learning neural network 126 identifying an image of a stop sign, As another example, the controller 132 may provide control signals to change direction and/or speed to avoid a rear-end collision responsive to cooperative learning neural network 126 identifying a precursor to a rear-end collision.

[ 048 ] Figure 2 is a flowchart illustrating a method arranged in accordance with examples described herein. Method 200 includes block 202, block 204, block 206, block 208, and block 210. In block 202, data may be received from one or more other vehicle(s). to block 204, which may follow block 202, a cooperative Learning neural network may be trained with data from these related vehicles such as Vehicle 102 and/or Vehicle 104 of Figure 1. The block 206 may follow block 204. In block 206, weights for the cooperative learning neural network may be stored. Block 208 may follow block 206. In block 208, a crash condition may be identified using the cooperative learning neural network, Block 210 may follow block 208. In block 210, operation of a vehicle may be modified using the cooperative learning neural network, Additional, fewer, and/or otherwise ordered blocks may be used in other examples. The method 200 may be performed, for example, by vehicles described herein, such as by vehicle 102 and/or vehicle 104 of Figure .1 ,

[049 ] In block 202, datu may be received from other vehicle(s). For example referring hack to Figure 1 , data from vehicle 104 may be received at vehicle 102. Data may be wirelessly transmitted from one vehicle to another in examples described herein, such as by utilizing 5G wireless communication techniques. Utilizing 5G wireless communication techniques may ailow for sufficient speed and/or bandwidth of data transfer to allow data received front other vehicles to be timely used in controlling a vehicle, Other, legacy communication schemes or protocols thut do not support high bandwidth and/or low latency data transfers may be insufficient to transfer enough or timely data to support the techniques described herein.

[050] Data may be received from any number of vehicles including 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 50, 100, 500, 1000 vehicles, or another number of vehicles in other examples, Data received may represent data generated by sensors of the other vehicles. For example, image data collected from image sensors (e.g., cameras) located on other vehicles may be received in block 202. Other data which may be collected in block 202 includes, but is not limited to, vehicle identity, location, speed, state, destination, or other vehicle, component, or environmentul parameters. In some examples, data received in block 202 may be data dircclly for use by a cooperative learning neural network (e.g., weights for a neural network may be sent from another vehicle and received in block 202), The operations of block 202 may, for example, be performed by u receiving component in a system or subsystem of a vehicle, such as transceiver 1 10 or transceiver 120.

[051] In block 204, data from other vehicles may be used to train a cooperative learning neural network, such as the cooperative learning neural network 128 and/or cooperative Learning neural network 126 of Figure 1. Data from other vehicles (e.g., data received in block 202) may be combined with data from a vehicle in which the cooperative learning neural network is running in block 204, For example, data may be generated by one or more sensors of the vehicle in which the cooperative learning neural network is operating, The generated data may be used in block 204 together with the data received in block 202 to train the cooperative learning neural network, .Both data from the local vehicle and data from other vehicte(s) may be used to train a cooperative learning neural network in block 204.

[052 ] Generally, a neural network, including cooperative Learning neural networks described herein, may he considered as an electronic network of nodes. A set of i nput vulues may be provided lo the neural network and associated weights may be implemented by the nodes. Λ function may be used to combine the weights and the input values and map the results lo an output. In this manner, based on the output, a netiral network may classify the output. Training (e.g., learning) a neural network generally refers to iterative!/ updating the weights (e.g., parameters) using a rule. Training may utilize two or three kinds of variables in some examples, the input signals (e.g„ training samples), targeted outputs (e.g., supervised samples), and initial weights (e.g., all zero, randomly selected, or other initial weight selection).

[ 053] Training may involve supervised training ··· e.g., comparing the classification of the input values with a known actual classification of the input values, Errors from an initial classification of the input values may be fed back into (he neural network, and used to modify the network (e.g., the weights and/or the function to map to the output) (he next time around. Many iterations of adjustment may occur during training. Accordingly, supervised training (e.g., learning) may refer to oil three kinds of variables (eg,, input signals, targeted outputs, and initial weights) are used to update the weights. Examples of supervised training methodologies may include linear regression,, random forest, and support vector machines. In some examples, training may involve unsupervised training - e.g., using a methodology to model the underlying structure or distribution in the input: data in order to classify the data. Unsupervised learning (e.g., self-learning) may refer to the use of input signals and initial weights, without the use of targeted outputs, to update the weights of the neural network, for example, no targeted output may be provided in unsupervised learning examples. Examples of unsupervised Learning methodologies include k-tntans methodologies and a priori methodologies.

[054] Cooperative leaning neural networks described herein generally refer to neural networks where the input signals and/or targeted outputs may come from different sources or different objects, For example, some of the input signals and/or targeted outputs used may be communicated to the cooperative learning neural networks over a wireless communications network (e.g., a 50 communications network).

[ 055] 50 communication systems may allow cooperative learning neural networks described herein to train using sensor data communicated over the SG communication system. Generally, to be usable in a real-time setting, it may not be feasible without a high bandwidth 50 communication system to make use of sensor data communicated from another device. By real-time, herein generally may refer to occurring within a threshold time that is based at least in part on the condition identified. For example, realtime may refer to occurring with an amount of time where an action may be taken to avoid a crash condition described herein. By allowing for higher bandwidth communication of sensor data, 5G conummication systems described herein may facilitate cooperative learning neural networks which may make use of sensor data communicated over the 5G communication system,

[056] As a result of the training, the cooperative learning neural network may develop sets of weights, each of which correspond to an identifiable condition (e.g., an object in an image and/or a precursor to a crash condition). The weights developed as a result of training in block 204 may allow the cooperative learning neural network to more accurately identify a condition (e.g., an object in an image and/or a precursor to a crash condition) than training based solely on sensor data from the single vehicle in which the cooperative learning neural network is operating. In some examples, training in block 204 may be unsupervised training, in some examples, training in block 204 may be supervised training (e.g., training which occurs with reference to one or more known objects and/or conditions), The operations of block 204 may, for example, be performed by a cooperative learning neural network in a system or subsystem of a vehicle, such as cooperative learning neural network 126 und/or cooperative learning neural network 128 of Figure 1 .

[057] In some examples, one or more criteria may be used to identify vehicles from which inputs may be used in block 204. For example, vehicles who may have data relevant to a condition and/or image to be identified may be used in block 204. Considering the case of a cooperative learning neural network which may identify a precursor to a rear-end collision, for example, data from one or more vehicles behind a vehicle receiving the data may be used to train in block 20A A location of various vehicles may be provided to a vehicle and the location of the vehicle may be used to determine whether other data from that vehicle will be used in training in block 204. For example, data from vehicles that are within u threshold distance of the receiving vehicle may be used in some examples. Considering the case of a cooperative learning neural network which may identify an object from an image, data from vehicles in a location which are likely to have a view of the object may he used in block 204,

[058 ] In block 206, weights may be stored for use by a cooperative learning neural network. The weights may be the weights calculated during the training in block 204 and.or may be the weights which may be received from other vehicles. Weights may be stored in memory, e.g., memory 124 and/or memory 1 14 of Figure 1, The operations of block 206 may, for example, be performed by a cooperative learning neural network in a system or subsystem of a vehicle, such as cooperative learning neural network 126 and/or cooperative learning neural network 128 of Figure 1 .

[059] In block 208, during operation of a vehicle implementing a cooperative learning neural network, a precursor to a crash condition may be identified using the cooperative learning neural network. For example, input data from multiple vehicles may be input to A cooperative learning neural network. The cooperative learning neural network may analyze the input data using the stored weights, Based on the analysis, a condition may be identified. For example, the set of weights may be identified by the cooperative learning neural network which produce a particular output (e.g., a maximum output, a minimum output, or a predetermined output) responsive to inputs received during use. The set of weights may be indicative of a condition (e.g., a precursor to a crash condition), in this manner, me condition may be identified by the cooperative learning neural network. The operations of block 204 may, for example, be performed by a cooperative learning neural network in a system or subsystem of a vehicle, such as cooperative learning neural network 126 and/or cooperative learning neural network 128 of Figure 1.

[060| In some examples, an object in image data provided to the cooperative learning neural network may be identified instead of or in addition to block 208. For example, input data representing images from multiple vehicles may be inputted to a cooperative learning neural network, The cooperative learning neural network may analyze the input data using the stored weights, Based on (he analysis, an object in the image data may be identified. For example, the set of weights may be identified by the cooperative learning neural network which produce a particular output (e.g., a maximum output, a minimum output, or a predetermined output) responsive to inputs received during use. The set of weights may be indicative of an object (e.g., a stop sign, street sign, person, animal, building). In this manner, the object may be identified by the cooperative learning neural network.

[061 ] In block 210, operation of a vehicle may be modified using the cooperative learning neural network. For example, the cooperative learning neural network may identify an object in an linage and/or a precursor to a crash condition. Responsive to the identification, a controller in the vehicle ruiuung the cooperative learning neurit 1 network and/or in another vehicle may take action to modify operation of the vehicle. For example, responsive to identification of a stop sign in an image, the vehicle may be controlled to slow and/or stop, Responsive to identification of a precursor to a crash condition, the vehicle may change speed, direction, or other parameter, to avoid the crash condition. The operations of black 204 may. for example, be. performed by a cooperative learning neural network in a system or subsystem of a vehicle, such us cooperative learning neural network 126 and/or cooperative learning neural network. 128 of Figure 1 [062] Figure 3 is a schematic illustration of an example system arranged in accordance with examples described herein, The system 300 includes satellite 302, monitoring infrastructure 304, monitoring infrastructure 306, computing system 308, vehicle 310, vehicle 3 ( 2, vehicle 314, vehicle 316, vehicle 31 8, and vehicle 320, Other, fewer, and/or different components may be used in other examples, Any of the vehicles 310*320 may he implemented by and/or used to implement vehicles described herein, .such as vehicle 102 and vehicle 104 ofFigure 1. Any of the vehicles 310-320 may implement cooperative learning neural networks as described herein, such as with reference to Figure 2.

[063] Figure 3 illustrates lhal data generated by sensors on vehicles in systems described herein may be communicated to a variety of systems and devices. In some examples, data generated by any of the vehicles 310-320 may be transmitted directly to uny of the other vehicles, e.g., using transceivers included in the vehicles as described herein. Ln some examples, data generated by any of the vehicles 310-320 may be communicated to other vehicles and/or systems (e.g., computing system 308) using communications infrastructure, such as monitoring infrastructure 304, monitoring infrastructure 306, and/or satellite 302. In some examples, monitoring infrastructure 304, monitoring infrastructure 306, and/or satellite 302 may be included in 5G communications infrastructure. In some examples, data may be received by any of the vehicles 310*320 using communications infrastructure such as monitoring infrastructure 304, monitoring infrastructure 306, and/or satellite 302. Other computing systems, such as computing system 308 may provide data for transmission to any of the vehicles 310* 320, which data may be communicated, e.g.,, using monitoring infrastructure 304, monitoring infrastructure 306, and/or satellite 302.

[064] Data may initially be obtained with one or more sensors associated with the vehicles 310-320. The sensor data may itself be referred to as "raw data" ··· e.g., data which may be captured by a sensor without additional processing, compression, conversion, scaling, or the like. One or more subsystems described herein may process the raw data (e.g., compress and/or evaluate the raw data) to provide processed data.

[065 ] Accordingly, in some examples, the vehicles 310-320 may transmit data among one another, T he data communicated may include any sensor data (e.g., speed, locution, environmental condition, component status nnd-'or condition, direction). 5G communication techniques may be used to transmit and receive data among vehicles described herein. In some examples, alt or portions of raw sensor data (e.g., unprocessed camera captured-signals and sensor data) may be communicated between vehicles. In this manner, one vehicle may utilize raw sensor data from another vehicle in a same manner as though the data originated from a sensor of the first vehicle. In some examples, the use of 50 communications techniques may provide sufficient speed and/or bandwidth to make the transmission of raw sensor data sufficiently timely that the raw data may be used to timely make adjustments to vehicle operation.

[066] Vehicles 310*320 may transmit date among and between one another regarding object!! and/or conditions identified using cooperative learning neural networks described herein. The data may be unicast and-or broadcast. In some examples, a vehicle may transmit data with an identified recipient (e.g., another particular vehicle). In some examples, a vehicle may broadcast and/or multicast data without, or in addition to, identifying a particular recipient. Objects in the environment, surroundings, and/or traffic may be communicated from one vehicle to another. 5G communication techniques may be used to transmit and receive data among vehicles described herein.

[ 067 ] Vehicles 310-320 may transmit all or portions of datu (e.g., speed, location, environmental condition, component status and/or condition, direction) to one or more remote computing systems, such as computing system 308, 5G communication techniques may be used to communicate with the remote computing system. The remote computing system may be implemented using, for example, a cloud-based computing system (e.g., a remote server, a computing system of a wireless service provider), and/or computing resources made available through one or more service providers. In an antilogous manner, vehicles 310-320 may transmit data regarding objects and/or conditions identified using cooperative learning neural networks described herein to remote computing systems, such as computing system 308. For example, objects in the environment, surroundings, and/or traffic may be communicated from a vehicle to a remote computing system.

[068] In some examples, all or portions of raw sensor data (e.g., unprocessed camera captured-signals and sensor data) may be communicated between a vehicle and a remote computing system, such as computing system 30S. In some examples, the use of 5G communications techniques may provide sufficient speed and/or bandwidth to make (he transmission of raw sensor data sufficiently timely that the raw data, may be used to timely make adjustments to vehicle operation.

[069] In this manner, cooperative learning neural networks may be used to identify objects, e.g., obstacle 322 which may be one or more people along a roadway. Other obstacles may also be identified, such as but not limited to, debris, road kill, buildings, animals, or combinations thereof. For example, data from vehicle 314 and vehicle 316 may be used to train a cooperative learning neural network to identify obstacle 322.

[070] Cooperative learning neural networks described herein may also be used to detect precursors to collision conditions, such as a passible collision between vehicle 310 and vehicle 312 depicted in Figure 3. Data from vehicle 310 and vehicle 312 may be used to train a cooperative learning neural network in vehicle 310 and/or vehicle 312 to identity precursors to the possible collision. A possibility of a rear-end collision between vehicle 316 and vehicle 320 may be monitored by a cooperative learning neural network in vehicle 316 and/or vehicle 320. For example, data from both vehicle 316 and vehicle 320 may be used to train a cooperative learning neural network in vehicle 316 and/or vehicle 320 to identify a precursor to a rear-end collision condition,

[071 ] Other conditions and/or objects may also be identified, such as a lane change by one or more vehicles. A lane change of vehicle 314 is depicted in Figure 3.

[072| Figure 4 is a schematic illustration of a cooperative learning neural network urranged in accordance with examples described herein. For example, the cooperative learning neural network 400 may be used to implement and/or may be implemented by cooperative learning neural networks described herein including cooperative learning neural network 128 and/or cooperative learning neural network 126 of Figure 1.

[073] The cooperative learning neural network 400 includes an input layer including a number of input variables. The cooperative learning neural network 400 includes a hidden layer including a number of hidden layer vnriables zj. The cooperative learning neural network 400 includes an output layer including a number of output layer variables z'k. Generally, the cooperative learning neural network 400 operates by combining inputs from the input layer in weighted combinations vising weights wij to calculate hidden layer variables zi. The hidden layer variables ure combined in weighted combinations using weights w'jk to calculate output layer variables A. While a single hidden layer is shown in Figure 4, tiny number of hidden layers may be used in cooperative learning neural networks described herein.

[074] The weights wn and wjk may be used to implement and/or may he implemented by weights described herein, such as weights 106 and-or weights 130 of Figure 1. Λ constraint may be evaluated against the output layer variables %\ to identity an object and/or condition described herein. For example, a function of the output layer variables may be evaluated, and the weights which minimize, maximize, or produce a predetermined value of that function, may be selected.. The condition and/or object associated with those weights may (hen be identified as present in the input.

[075 ] For example, different conditions and/or objects may make the weights converge to different values. For instance, for a "stop" sign, the values of the first weight may converge to 0.5, the value of the second weight may be 0.3, the rest of the weights may he zero, However, for a different object, such as a building, the value of the third weight may be 0.6 but the rest of the weights may be 0.5. Examples of cooperative neural networks described herein may provide more accurate weights to identify each object and/or condition than using data from each vehicle separately.

[076] As described herein, inputs from multiple vehicles may be used as input to a cooperative learning neural network. Suppose that x1(n), x2(n), . . . xM(n) and yi(n), . . . yN(n) are input vectors of two vehicles (e.g., vehicle 102 uud vehicle 104 of Figure 1) at lime n. The input vectors may be supplied as (he variables of the input layer shown in Figure 4. The input vectors may contain values representing any of a variety of values, such as pixel intensity and'or color values of pixels in an linage, and/or one or more sensor values,

[077] A weighted function, an II matrix, may represent a transfer function of variables common to the input vectors used as inputs to the cooperative learning neural network (e.g., of the multiple vehicles used as inputs). In this manner, the outputs may be determined based on the weighted function in conjunction with weights which ate specific to (e.g., associated with) the predicted output specific to each of the Input vehicles, in this manner, the weights and predicted outputs of the cooperative learning neural network 400, U and V vectors, may be represented mathematically as follows:

[0781 where y is a learning rate. Note that γ and/or H may be b minded by unity in some examples.

[0791 Eq.( 1) is an examples representative of unsupervised learning and cooperative learning where the weights W(ivi- 1) may be totally calculated by the initial weights W(0) and input signals (e.g., the first input x(n) from vehicle X and the second input from the vehicle Y).

[0801 In some examples, the cooperative learning neural network may utilize unsupervised learning. Unsupervised learning may refer to learning that docs not rely on comparison to reference images und/or reference data from known conditions, The unsupervised learning may have a large dynamic range and'or take a lengthy amount of time for convergence. This may increase the complexity of hardware used to implement the cooperative teaming neural network and/or time taken to implement the cooperative learning neural network. For example, the value of w{ 1 ,n), w(2,n), etc. may be very large and/or very small, which may necessitate a larger hardware resource than desired. In some examples, a norm-bounded teaming algorithm may be used, which may be expressed as follows:

[081 ] such that

[082] In this manner, the value of the weights which may be provided by the cooperative learning neural network arc bounded by the initial weights,

[083 ] If the weights have a very large dynamic value range, more bits (e.g., 64 or 32 bits) operations or even floating-point operations may be needed tor each computation step in order to provide sufficient precision. However, if the dynamic range is small, e.g. , 16 bit. or 8 bit, smaller operations may be sufficient. By adding one more multiplication, examples of the norm-bounded algorithm described, herein may perform the normalization processing in each step of the training so that the value of each weight may be limited to n normalized range.

[084| FIG. 5 illustrates an example of fl wireless communications system 500 in accordance with aspects of the present disclosure, The wireless communications system 500 includes a base station 510, mobile devices 515, a drone (or unmanned aerial vehicle) 517, n small cell 530, and vehicles 540, 545. Vehicles 540, 545 may be used to implement and/or may be implemented by vehicles described herein, such as the vehicles 102 and/or 104 of FIG. 1. For example, vehicles 540 and 545 may include a cooperative learning neural network as described herein, in some examples, mobile devices 515 and or drone 51 ? include cooperative learning neural networks.

[085] The base station 510 and small ceil 530 may be connected to a network that provides access to the Internet and traditional communication links. The system 500 may facilitate a wide-range of wireless communications connections in a 5G system that may include various frequency bands, including hut not limited to: a sub-6 GHz band (e.g., 700 MHz: communication frequency), mid-range communication bands (e.g„ 2.4 GH2), and mmWave bands (e.g. 24 GHz). Additionally or alternatively, the wireless communications connections may support various modulation schemes, including but not limited to: filter bunk multi-carrier (FBMC), the generalized frequency division multiplexing (CFDM), universal filtered multi-carrier (UFMC) transmission, bi-orthogonal frequency division multiplexing (BFDM), sparse code multiple access (SCMA), non-orthogonal multiple access (NOMA), multi-user shared access (MtJSA), and laster-than-Nyquist (FTN) signaling with time-frequency packing. Such frequency bands and modulation techniques may be a part of a standards framework, such as New Radio (NR) Long Term Evolution (LTE) or other technical specification published by an organization like 3GPP or IEEE, which may include various specifications lor subcurrier frequency ranges, a number of tubcurriers, uplink/downlink ttansmisjiion .speeds, TDD/FDD, and/or other aspects of wireless communication protocols.

[086] The system 500 may depict aspects of a radio access network (RAN), and system 500 may be in communication with or include a core network (not shown). The core network may include one or more serving gateways, mobility management entities, home subscriber servers, and packet data gateways, The core network, may facilitate user and control plane links to mobile devices via the RAN, and it may be an interface to on externa! network (e.g., the Internet), Base stations 510, communication devices 520, and small cells 530 may be coupled with the core network or with one another, or both, via wired or wireless backhaul links (e.g. , S1 interface, X2 interface, etc.),

[087] The system 500 may provide communication links with devices or "things," such as sensor devices, e.g., solar cells 537, to provide on Internet of Things ("loT") framework. Connected things within the lo.V may operate within frequency bands licensed to and controlled by cellular network service providers, or such devices or things may. Such frequency bands and operation may be referred to as narrowband loT (NB- loT) because the frequency bands allocated for loT operation may be small or narrow relative to the overall system bandwidth. Frequency bands allocated for NB-!oT may have bundwidths of 1, 5, 10, or 20 MHz, for example,

[ 088] Additionally or alternatively, the loT may include devices or things operating at different frequencies than traditional cellular technology to facilitate use of the wireless spectrum. For example, an loT framework may allow multiple devices in system 500 to operate at a sub-6 GHz baud or other industrial, scientific, and medical (ISM) radio bands where devices may operate on a shared spectrum for unlicensed uses.

[089 ] The sub-6 GHz band may also be characterized as an NB-loT band, for example, in operating at low frequency ranges, devices providing sensor data for "tilings " such as solar cells 537, may utilize less energy, resulting in power-efficiency and may utilize less complex signaling frameworks, such that devices may transmit asynchronously on thut sub-fi GHz band. The sub-6 GHz band may support a wide variety of use cases, including the communication of sensor data from various sensors devices. Examples of sensor devices include sensors for detecting energy, heat, light, vibration, biological signals, distance, speed, acceleration, or combinations thereof. Sensor devices may be deployed on buildings, individuals, and/or in other locations in the environment. The sensor devices may communicate with one another and with computing systems which may aggregate and/Or analyze the data provided from one or multiple sensor devices in the environment.

[090] in such a 5G framework, devices may perform functionalities performed by base stations in other mobile networks (e.g., UMTS or LTE), such as forming a connection or managing mobility operations between nodes (e.g., handoff, paging, radio resource configuration reseleeiion etc.).

|091] By way of example, mobile device 515 may receive sensor data from the user utilizing the mobile device 515, and may transmit that sensor data on a narrowband loT frequency band to base station 510. In such an example, some parameters for the determination by the mobile device 515 may include availability of licensed spectrum, availability of unlicensed spectrum, and/or time-sensitive nature of sensor data, [092 ] Additionally or alternatively, mobile device 515 may form device-to-device (D2D) connections with other mobile devices or other elements of the system 500. For example, the mobile device 515 may form RFID, WiFi, MulliFire, Bluetooth, or Zigbee connections with other devices, including communication device 520 or vehicle 545. in some examples, D2D connections may be made using licensed spectrum bands, and such connections may be manured by a cellular network or service provider. Accordingly, while the above example was described in the context of narrowband loT, it can be appreciated that other deviee-to-device connections may be utilized by mobile device 515 to provide information (e.g., sensor data) collected on different frequency bands than a frequency band determined by mobile device 515 for cransmission of that information.

[093 ] Moreover, some communication devices may facilitate ad-hoc networks, for example, u network being formed with communication devices 520 attached to stationary objects (e.g., lampposts in Fig, 5) and the vehicles 540, 545, without a traditional connection to a base station 510 and'or a core network necessarily being formed. Other stationary objects may be used to support communication devices 520, such as, but not limited to, trees, plants, posts, buildings, blimps, dirigibles, balloons, street signs, mailboxes, or combinations thereof. In such a system 500, communication devices 520 and small cell 530 (e.g., a picooell, femtoceU, WLAN access pomt, cellular hotspot, etc.) may be mounted upon or adhered to another structure, such as lampposts and buildings to facilitate the formation of ud-hoc networks and other loT-based networks. Such networks may operate at different frequency bands than existing technologies, such as mobile device 515 communicating with base station 510 on a cellular communication band.

[094] The communication devices 520 may form wireless networks, operating in either a hierarchal or ad-hoc network fashion, depending, in part, on the connection to another element of the system 500, For example, the communication devices 520 may utilize a 700 MHz communication frequency to form n connection with the mobile device 515 in an unlicensed spectrum, while utilizing a licensed spectrum communication frequency to form another connection with the vehicle 545. Communication devices 520 may communicate with vehicle 545 on a licensed spectrum to provide direct access for time-sensitive data, for example, data for an autonomous driving capability of the vehicle 545 on a 5.9 GHz band of Dedicated Short Range Communications (DSRC). [095 ] Vehicles 540 and 545 may form an ad- hoc network at a different frequency batid than the connection between the communication device 520 and the vehicle 545. For example, for a high bandwidth connection to provide time-sensitive data between vehicles 540, 545, a 24 GHz mmWave band may be utilized for transmissions of data between vehicles 540, 545, For example, vehicles 540, 545 may share real-lime directional and navigation data with each other over the connection while the vehicles 540, 545 pass each other across a narrow intersection line. Each vehicle 540, 545 may be tracking the intersection line and providing image data to an image processing algorithm to facilitate autonomous navigation of each vehicle while each travels along the intersection line. In some examples, this real-time data may also be substantially simultaneously shared over an exclusive, licensed spectrum connection between the communication device 520 and the vehicle 545, for example, for processing of image data received at both vehicle 545 and vehicle 540, as transmitted by the vehicle 540 to vehicle 545 over the 24 GHz mmWave band, While shown as automobiles in Fig. 5, other vehicles may be used including, but not limited to, aircraft, spacecraft, balloons, blimps, dirigible*, trains, submarines, boats, ferries, cruise ships, helicopters, motorcycles, bicycles, drones, or combinations thereof.

[096| While described in the context of a 24 GHz mmWave band, it can be appreciated that connections may be formed in the system 500 in other mmWave bands or other frequency bands, such as 28 GHz, 37 GHz, 38 GHz, 39 GHz, which may be licensed or unlicensed bands. In some cases, vehicles 540, 545 may share the frequency band that they are communicating on with other vehicles in a different network. For example, a fleet of vehicles may pass vehicle 540 and, temporarily, share the 24 GHz mmWave bund to form connections among that fleet, in addition to the 24 GHz mmWave connection between vehicles 540, 545. As another example, communication device 520 may substantially simultaneously maintain a 700 MHz connection with rhe mobile device 515 operated by a user (e.g., a pedestrian walking along the street) to provide information regarding a location of the user to the vehicle 545 over the 5.9 Gl lz hand. In providing such information, communication device 520 may leverage antenna diversity schemes us part of a massive MIMO framework, to facilitate lime-sensitive, separate connections with both the mobile device 515 and the vehicle 545. A massive MIMO framework may involve a transmitting and/or receiving devices with a large number of antennas (e.g., 12, 20, 64, 128, etc.). which may facilitate precise bcamforming or spatial diversity unattainable with devices operating with fewer amentias according to legacy protocol:; (e.g., WiFi or LTE).

[097 ] The base station 510 and small cell 530 may wirelcssly comnuuiicate with devices in the system 500 or other communication-capable devices in (he system 500 having ut the leust a sensor wireless network, such us solar cells 537 that may operate on an active/sleep cycle, and/or one or more other sensor devices. The base station 510 may provide wireless communications coverage for devices that enter its coverages area, such as the mobile device 515 and the drone 517. The small cell 530 may provide wireless communications coverage for devices that eater its coverage area, such as near the building that the small cell 530 is mounted upon, such as vehicle 545 and drone 517,

[098] Generally, a small cell 530 may be referred to us a picocell and provide coverage for a local geographic region, for example, coverage of 200 meters or less in some examples, This may be contrasted with at macrocell, which may provide coverage over a wide or targe area on the order of several square miles or kilometers. In some examples, a small cell 530 may be deployed (e.g., mounted on a building) within some coveruge areas of a base station 510 (e.g., a macrocell) where wireless communications traffic may he dense according to a traffic analysis of that coverage area. For example, a small cell 530 muy be deployed on the building in Fig, 5 in the coverage area of the base station 510 if the base station 510 generally receives and/or transmits a higher amount of wireless communication transmissions than other coverage areas of that base station 510. A base station 510 may be deployed in a geographic area to provide wireless coverage for portions of thai geographic area. As wireless communications traffic becomes more dense, additional base .stations 510 may be deployed in certain areas, which may alter the coverage area of an existing base station 510, or other support stations may be deployed, such as a small cell 530. Small ceil 530 may be a femtocell, which may provide coverage for an area smaller than a picocell (e.g., 100 meters or less in some examples (e.g., one story of a build ing)),

[099] While base station 510 and small cell 530 may provide communication coverage for a portion of the geographical area surrounding their respective areas, both may change aspects of their coverage to facilitate faster wireless connections for certain devices. For example, the small cell 530 may primarily provide coverage for devices surrounding or in the building upon which the small cell 530 is mounted. However, the small cell 530 may also delect LhaL a device has entered is coverage area and adjust its coverage area to facilitate a faster connection to that device.

[0 100] For example, a small ceil 530 may support a massive M1MO connection with the drone 517 and, when the vehicle 515 (inters it coverage arc*, the small cell 530 adjusts some antennas to point directionuHy in a direction of the vehicle 545, rather thun the drone 517, to facilitate a massive MIMO connection with the vehicle, in addition to the drone 517, in adjusting some of the antennas, the small cell 530 may not support, as fast as a connection to the drone 5.17, as it had before the adjustment. However, the drone 517 may also request a connection with another device (e.g., base station 510) in its coverage area that may facilitate a similar connection as described with reference to the small cell 530, or u different (e.g., faster, more reliable) connection with the base station 510, Accordingly, the system 530 may enhance existing communication links in providing additional connections to devices that may utilize or demand such links. For example, the small cell 530 may include a massive MIMO system that directionally augments a link to vehicle 545, with antennas of the small cell directed to the vehicle 545 for a specific time period, rather than facilitating other connections {e.g., the small cell 530 connections to the base station 510, drone 517, or solar cells 537). In some examples, drone 5.17 may serve as a movable or aerial base station,

[0 101] The wireless communications system 500 may include devices such as base station 510, communication device 520, and small cell 530 that may support several connections to devices in the system 500, Such devices may operate in a hierarchal mode or an ad-hoc mode with other devices in the network of system 500. While described in the context of a base station 510, communication device 520, and small cell 530, it can be appreciated that other devices that can support several connections with devices in the network may be included in system 500, including but. not limited to: macrocells. femtocells, routers, satellites, and RFID detectors.

[0 102] The wireless communication system 500 may suppoit higher bandwidth and/or lower latency communications with the techniques described above. Devices with system 500 may therefore leverage such techniques to exchange data for the benefit of training and utilizing cooperative learning neural networks, as described with reference to the preceding figures. [0103] Certain details tire set forth above to provide a sufficient understanding of described embodiments. However, it will be clear to one skilled in the art that embodiments may be practiced without various of these particular details. The description herein, in connection, with the appended drawings, describes example configurations and does not represent all the examples that may be implemented or that are within the scope of the claims. The terms "exemplary" and "example" as may be used herein means "serving as an example, instance, or illustration." and not "preferred" or "advantageous over other examples." The detailed description includes specific details for the purpose of providing an understanding of the described techniques. These techniques, however, may be practiced without these specific details, hi some instances, well-known structures and devices are shown in block diagram form in order to avoid obscuring the concepts of the described examples.

[0104 ] Information and signals described herein may be represented using any of a variety of different technologies and techniques. For example, data, instructions, commands, information, signals, bits, symbols, and chips that may be referenced throughout the above description may be represented by voltages, currents, electromagnetic waves, magnetic fields or particles, optical fields or particles, or any combination thereof.

[ 0105] Techniques described herein may be used for various wireless communications systems, which may include multiple access cellular communication systems, und which may employ code division multiple access (CDMA), time division multiple access (TDMA). frequency division multiple access (FDMA), orthogonal frequency division multiple access (OFDMA), or single carrier frequency division multiple access (SC- FDMA), or any a combination of such techniques. Some of these techniques have been adopted in or relate to standardized wireless communication protocols by organizations such as Third. Generation Partnership Project (3GPP), Third Generation Partnership Project 2 (3GPP2) and IEEE. These wireless standards include Ultra Mobile Broadband (UMB), Universal Mobile Telecommunications System (UMTS), Long Term Evolution (LTli), LTE-Advanced (LTE-A), LTE-A Pro, New Radio (NR), IEEE 802.1 1 (WiFi), and IRER 802.16 (WiMAX), among others.

[0106] The terms "50" or "50 communications system" may refer to system* that operate according to standardized protocols developed or discussed after, for example, LTE Releases 13 or .14 or WiMAX 802.16c-2005 by their respective sponsoring organizations. The features described herein may be employed in systems configured according to other generations of wireless communication systems, including those configured according to the standards described above,

[0107] The various illustrative blocks and modules described in connection with the disclosure herein may he implemented or performed with a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field- programmable gate array (FPGA), or other programmable logic device, discrcre gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A general-purpose processor may be a microprocessor, but in the alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. A processor may also be implemented as a combination of computing devices a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration).

[ 0108] The functions described herein may be implemented in hardware, software executed by o processor, firmware, or any combination thereof. If implemented in software executed by a processor, the functions may be stored on or transmitted over as one or more instructions or code on a computer-readable medium. Computer-readable media includes both non-transitory computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. A non-transitory storage medium may be any available medium that can be accessed by a general purpose or special purpose computer. By way of example, and not limitation, non-transitory computer-readable media can comprise RAM, ROM, electrically erasable programmable read only memory (EEPROM), or optica! disk storage, magnetic disk storage or other magnetic storage devices, or any other non- transitory medium that can be used to cany or store desired program code means in the form of instructions or data structures and that can be accessed by a general-purpose or special-purpose computer, or a general -purpose or special-purpose processor.

[109] Also, any connection is properly termed a computer-readable medium, For example, if the software is transmitted from a website, server, or other remote source using a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of medium. Combinations of the above are ulso included within the scope of computer-readable medio.

[0110] Other examples and implementations are within the scope of the disclosure and appended claims. For example, due to the nature of software, functions described above can be implemented using software executed by a processor, hardware, firmware, hardwiring, or combinations of any of these. Features implementing functions may also he physically located at various positions,, including being distributed such that portions of functions are implemented at different physical locations.

[0111] Also, as used herein, including in the claims, "or" as used in a list of items (for example, a list of items prefaced by a phrase such as "at least one οΓ' or "one or more of) indicates an inclusive list such that, for example, a list of at least one of A, B, or C means A or B or C or AB or AC or BC or ABC (i.e. , A and B and C). Also, as used herein, the phrase "based on" shall not be construed as a reference to a. closed set of conditions. For exumple, an exemplary step that is described as "based on condition A" may be based on both a condition A and n condition B without departing from the scope of the present disclosure. In other words, as used herein, the phrase "based on" shall be construed in the same manner as the phrase "based at least in part on."

[0112] From the foregoing it will be appreciated that, although specific embodiments have been described herein for purposes of illustration, various modifications may be made while remaining with the scope of the claimed technology, The description herein is provided to enable a person skilled in the art to make or use the disclosure, Various modifications to the disclosure will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other variations without departing from the scope of the disclosure. T hus, the disclosure is not limited to the examples and designs described herein, but is to be accorded the broadest scope consistent with the principles and novel features disclosed herein,