Qualcomm Incorporated

Alphaics Corporation

Qpicloud Technologies Private Limited

Qpiai India Private Limited

Nagendra Nagaraja

Khaled Helmi El-Maleh

Eddie L T Choy

Gavin Bernard Horn

Miguel Griot

Sachin Kumar

Saiyiu Duncan Ho

Lakshya Priyadarshi

Abhinand Balachandran

Aswanth Krishnan

Systems and methods for optimizing the placement of experts in a mixture-of-experts (MoE) model are disclosed. A system can include one or more processors coupled to non-transitory memory. The system can obtain a set of attributes of a plurality of hardware components of a distributed computing environment. The system can determine a set of co-activation metrics for a plurality of experts of a mixture-of-experts model according to inference or training operations executed using the mixture-of-experts model. The system can generate a quadratic unconstrained binary optimization (QUBO) data structure based at least on the set of attributes and the set of co-activation metrics. The system can assign the plurality of experts of the mixture-of-experts model to the plurality of hardware components of the distributed computing environment according to the QUBO data structure. The system can execute the mixture-of-experts model using the plurality of hardware components according to the assignment.

Systems and methods for optimizing mixture-of-experts model placement using QUBO-based bin packing

A system and method include receiving a resource headroom data of an edge device, receiving resource utilization data of a plurality of applications, and selecting a group of applications from the plurality of applications for installation on the edge device based upon the resource headroom of the edge device. The system and method also include computing a fitness score based upon a suitability of the selected group of applications for the edge device, generating a reward based on the fitness score, and using the reward to refine the selection of the group of selections in subsequent iterations.

Profiling and application monitoring for edge devices based on headroom

A reinforcement learning processor specifically configured to train reinforcement learning agents in the AI systems by the way of implementing an application-specific instruction set is disclosed. The application-specific instruction set incorporates 'Single Instruction Multiple Agents (SIMA)' instructions. SIMA type instructions are specifically designed to be implemented simultaneously on a plurality of reinforcement learning agents which interact with corresponding reinforcement learning environments. The SIMA type instructions are specifically configured to receive either a reinforcement learning agent ID or a reinforcement learning environment ID as the operand. The reinforcement learning processor is designed for parallelism in reinforcement learning operations. The reinforcement learning processor executing of a plurality of threads associated with an operation or task in parallel.

System and method for training artificial intelligence systems using a sima based processor

The embodiments herein disclose a system and method for implementing reinforcement learning agents using a reinforcement learning processor. An application-domain specific instruction set (ASI) for implementing reinforcement learning agents and reward functions is created. Further, instructions are created by including at least one of the reinforcement learning agent ID vectors, the reinforcement learning environment ID vectors, and length of vector as an operand. The reinforcement learning agent ID vectors and the reinforcement learning environment ID vectors are pointers to a base address of an operations memory. Further, at least one of said reinforcement learning agent ID vector and reinforcement learning environment ID vector is embedded into operations associated with the decoded instruction. The instructions retrieved by agent ID vector indexed operation are executed using a second processor, and applied onto a group of reinforcement learning agents. The operations defined by the instructions are stored in an operations storage memory.

Method and system for implementing reinforcement learning agent using reinforcement learning processor

System and method for training artificial intelligence systems using a SIMA based processor

The embodiments herein discloses a system and method for designing SoC by using a reinforcement learning processor. An SoC specification input is received and a plurality of domains and a plurality of subdomains is created using application specific instruction set to generate chip specific graph library. An interaction is initiated between the reinforcement learning agent and the reinforcement learning environment using the application specific instructions. Each of the SoC sub domains from the plurality of SoC sub domains is mapped to a combination of environment, rewards and actions by a second processor. Further, interaction of a plurality of agents is initiated with the reinforcement learning environment for a predefined number of times and further Q value, V value, R value, and A value is updated in the second memory module. Thereby, an optimal chip architecture for designing SoC is acquired using application-domain specific instruction set (ASI).

System and method for designing system on chip (SoC) circuits using single instruction multiple agent (SIMA) instructions

The embodiments herein discloses a system and method for designing SoC by synchronizing a hierarchy of SMDPs. Reinforcement Learning is done either hierarchically in several steps or in a single-step comprising environment, tasks, agents and experiments, to have access to SoC (System on a Chip) related information. The AI agent is configured to learn from the interaction and plan the implementation of a SoC circuit design. Q values generated for each domain and sub domain are stored in a hierarchical SMDP structure in a form of SMDP Q table in a big data database. An optimal chip architecture corresponding to a maximum Q value of a top level in the SMDP Q table is acquired and stored in a database for learning and inference. Desired SoC configuration is optimized and generated based on the optimal chip architecture and the generated chip specific graph library.

System and method for designing system on chip (SoC) circuits by synchronizing a hierarchy of SMDPs

The embodiments herein discloses a system and method for designing SoC using AI and Reinforcement Learning (RL) techniques. Reinforcement Learning is done either hierarchically in several steps or in a single-step comprising environment, tasks, agents and experiments, to have access to SoC (System on a Chip) related information. The AI agent is configured to learn from the interaction and plan the implementation of a SoC circuit design. Q values generated for each domain and sub domain are stored in a hierarchical SMDP structure in a form of SMDP Q table in a big data database. An optimal chip architecture corresponding to a maximum Q value of a top level in the SMDP Q table is acquired and stored in a database for learning and inference. Desired SoC configuration is optimized and generated based on the optimal chip architecture and the generated chip specific graph library.

System and method for designing system on chip (SoC) circuits through artificial intelligence and reinforcement learning

A reinforcement learning processor specifically configured to execute reinforcement learning operations by the way of implementing an application-specific instruction set is envisaged. The application-specific instruction set incorporates 'Single Instruction Multiple Agents (SIMA)' instructions. SIMA type instructions are specifically designed to be implemented simultaneously on a plurality of reinforcement learning agents which interact with corresponding reinforcement learning environments. The SIMA type instructions are specifically configured to receive either a reinforcement learning agent ID or a reinforcement learning environment ID as the operand. The reinforcement learning processor uses neural network data paths to communicate with a neural network which in turn uses the actions, state-value functions, Q-values and reward values generated by the reinforcement learning processor to approximate an optimal state-value function as well as an optimal reward function.

Processor for implementing reinforcement learning operations

Configuring a peer-to-peer (P2P) link in a multi-access wireless network includes receiving P2P configuration information from a base station at a UE supporting P2P communication. The first UE communicates directly with a second UE based on the P2P configuration information received from the base station. The first UE may send a configuration request message to the base station, and receive a responsive configuration message with the P2P configuration information from the base station, which messages may be Radio Resource Control (RRC) messages supporting P2P. In the alternative, the P2P configuration information may be provided in a system information block (SIB) broadcast by the base station. The P2P configuration information may indicate allocation of physical layer or Medium Access Control resources, or both allocated for P2P communication, and other information.

Configuration of user equipment for peer-to-peer communication

Techniques for supporting communication by a relay station are described. In an aspect, the relay station may support NACK Type 1 when operating in an amplify-and-forward (AF) mode. The relay station may receive a first transmission of a packet from an upstream station, determine PAPR of the first transmission, and send NACK Type 1 to the upstream station if high PAPR is detected. In another aspect, the relay station may support NACK Type 1 and NACK Type 2 when operating in a decode-and-forward (DF) mode. The relay station may perform PAPR decoding for the first transmission, send NACK Type 1 if PAPR decoding fails, perform channel decoding if PAPR decoding passes, and send NACK Type 2 to the upstream station if channel decoding fails. In yet another aspect, the relay station may operate in the AF mode or the DF mode.

Data transmission via a relay station with ACK/NACK feedback

A user computing device, such as mobile device, is provided with an electronic program guide (EPG). The EPG lists video content that is available or being broadcast on one or more websites. The video content may be considered to be premium content or user content. Premium content may be video content that is available from the website of a commercial broadcaster. Premium content may be generated by the commercial broadcaster. User content may be video content that has been generated and provided to websites for viewing and sharing by users such as members of the websites. An EPG may display listings of both user content and premium content.

Mapping mobile device electronic program guide to content

Aspects describe pairs of nodes negotiating air interfaces in a peer-to-peer and/or ad hoc network. The negotiation can take into account the capability of each node. The negotiation can be performed on a default link and after negotiation, the nodes can facilitate a communication handoff to the negotiated link. Further, one or more nodes can support multiple air interfaces, wherein nodes that support multiple air interfaces can establish more than one air interface with peer nodes over similar air interfaces or over different air interfaces as a function of negotiation between the nodes.

Air interface selection between nodes in peer-to-peer/ad-hoc networks

A method and apparatus for performing spectral allocation for a wireless link in a wireless communication system is disclosed. The method determines a distance/range for the wireless link and assigns an optimal sub-carrier frequency to the wireless link based on (1) a higher frequency to lower frequency priority and (2) on the value of the distance/range of the wireless link. Higher frequency sub-carriers are assigned to wireless links that have a shorter distance/range and lower frequency sub-carriers are assigned to wireless links that have a longer distance/range. The method of the present work increases the spectral efficiency of the wireless communication system and improves system bandwidth.

Ranging and distance based wireless link spectrum selection

Layered semantic security provides a high degree of security for a mobile device based upon contextual awareness that dynamically changes based upon interaction between a user and a near communication device, which in turn interacts with a network, which ultimately interacts to a far communication device. Generating a shared secret key with a master secret and changing contextual information based on context awareness provides immunity to chosen plain text attacks by providing semantic security at each layer. Thereby, relying upon the overall robustness of the layering of semantic security, processing and power resources consumed can be advantageously adjusted dynamically to enhance concurrent use and service life of a mobile communication device.

Growing community of inventors

Bengaluru, India

Nagendra Nagaraja