Hyperdimensional Computing for Low Size, Weight, and Power Systems

OUSD (R&E) CRITICAL TECHNOLOGY AREA(S): Trusted AI and Autonomy; Advanced Computing and Software; Human-Machine Interfaces; Space Technology The technology within this topic is restricted under the International Traffic in Arms Regulation (ITAR), 22 CFR Parts 120-130, which controls the export and import of defense-related material and services, including export of sensitive technical data, or the Export Administration Regulation (EAR), 15 CFR Parts 730-774, which controls dual use items. Offerors must disclose any proposed use of foreign nationals (FNs), their country(ies) of origin, the type of visa or work permit possessed, and the statement of work (SOW) tasks intended for accomplishment by the FN(s) in accordance with the Announcement. Offerors are advised foreign nationals proposed to perform on this topic may be restricted due to the technical data under US Export Control Laws. OBJECTIVE: Develop a unified hyperdimensional framework for integrating and optimizing the performance of heterogeneous artificial neural network (ANN) architectures to enable development of modular machine learning solutions. DESCRIPTION: Work on this topic should consider at least one of the following research areas. 1) Hyperdimensional Computing (HDC) Integration: Investigate and develop methods for seamlessly integrating diverse ANN architectures, such as convolutional neural networks (CNNs), recurrent neural networks (RNNs), and deep reinforcement learning (DRL) networks, into a single hyperdimensional processing pipeline. 2) Optimization and Resource Allocation: Research and develop optimization techniques that allow for dynamic resource allocation among the integrated ANNs, considering factors such as the computational requirements, data availability, and available resources. 3) Domain Adaptation and Transfer Learning: Explore how HDC can facilitate domain adaptation and transfer learning across different ANN architectures, ensuring that the unified framework remains flexible and adaptable to various DoD security applications and datasets. 4) Robustness and Fault Tolerance: Develop methods for integrating ANNs that provide inherent robustness and fault tolerance against noisy or incomplete data. 5) Energy Efficiency and Scalability: Optimize the integrated HDC framework for both energy efficiency and scalability, considering size, weight, and power (SWaP) constrained military platforms and applications that require processing large volumes of data. 6) Demonstration and Evaluation: Demonstrate the effectiveness of the unified HDC framework through evaluation on standard AI datasets, as well as custom experimentation tailored to DoD applications. PHASE I: Phase I awardee(s) will experiment with and assess feasibility of different approaches to implementing and optimizing hyperdimensional computing techniques for specific applications such as pattern recognition, sensor fusion, and information retrieval tasks. In support of this, awardee(s) will obtain baseline performance metrics, such as (but not limited to) accuracy, computational efficiency, robustness to noise, and scalability to evaluate the potential benefits of hyperdimensional computing over traditional methods. Based on these results, awardee(s) will identify and prioritize the most promising approaches for further development in Phase II, with the goal of advancing the state-of-the-art in hyperdimensional computing and creating new opportunities for practical applications. PHASE II: During Phase II, the awardee(s) will demonstration a prototype of a heterogenous machine learning system with an HDC backbone. Such demonstrable capabilities include can include (but are not limited to) efficient encoding and decoding, accuracy, computational efficiency, robustness to noise, and scalability. Prototype may involve integrating hardware and software components, such as specialized processors, memory systems, and software frameworks, to create a complete end-to-end solution. Lastly, awardees will identifying potential commercial applications. PHASE III DUAL USE APPLICATIONS: In Phase III, performers should expand ML capability demonstrations of heterogenous sensors on a single ML agent to orchestration of and/or collaboration with multiple agents via HDC as a shared information representation. The proposed technology is expected to start at TRL 5-6, concluding with a TRL 7 demonstration in a mission relevant environment, e.g. air, space, or ground. HDC stands to make artificial neural networks more human-interpretable and therefore lower the technical barrier to model adaptation by non-experts. Generally speaking, AI/ML as a data processing tool is considered a dual-use technology, since both industry and the AF desire ML solutions demonstrating high task performance, e.g. classification accuracy, from minimal training data. However, an SBIR is an appropriate vehicle to fund this type of research and development since the AF specifically has size, weight, and power (SWaP) restrictions not found in industry generally. HDC not only augments performance capabilities of artificial neural networks, it also drastically reduces the algorithmic resource requirements, making it suitable for ultra-low SWaP AF assets, e.g. UAVs and CubeSats. REFERENCES: 1. Kleyko, Denis, Dmitri A. Rachkovskij, Evgeny Osipov, and Abbas Rahimi. "A survey on hyperdimensional computing aka vector symbolic architectures, part i: Models and data transformations." ACM Computing Surveys 55, no. 6 (2022): 1-40. 2. Kleyko, Denis, Dmitri Rachkovskij, Evgeny Osipov, and Abbas Rahimi. "A survey on hyperdimensional computing aka vector symbolic architectures, part ii: Applications, cognitive models, and challenges." ACM Computing Surveys 55, no. 9 (2023): 1-52. KEYWORDS: hyperdimensional computing; vector symbolic architectures; holographic reduced representations; binary spatter codes; symbolic reasoning; edge processing; computational neuroscience; cognitive architectures; artificial neural networks; fusion