Call for abstracts

Artificial Intelligence (AI) is becoming ubiquitous and moving from the cloud directly into embedded devices. In both cases, the complexity of many AI problems requires substantial computational resources. For this reason, we have seen the emergence of dedicated hardware, particularly for that subset of AI that is referred to as deep learning. However, the range of AI techniques that could be employed in embedded applications is vast and it is not limited to deep learning. Autonomy applications such as driver-less cars and unmanned aerial vehicles span all phases of the classical Observe, Orient, Decide, and Act (OODA) loop. Architectures supporting these applications feature a situational assessment process that populates and maintains an internal world model, and a decision-making process that computes the best course of action to achieve the goals of the system, taking into account the state of the world. While deep learning has been successfully used to extract features from raw data such as images and to classify entities, advanced reasoning methods are used to assess complex situations and to make predictions. Similarly, while deep reinforcement learning shows promises for skill-level autonomy, other tasks such as mission planning, long-term planning, and contingency management rely on action models and planning algorithms that are based on logical representations like Markov Decision Processes (MDP), Partial Observable Markov Decision Processes (POMDP), and Hierarchical Task Networks (HTN).

Nowadays, reasoning and planning methods are implemented in software. The algorithmic complexity of these methods, especially when dealing with models of the world that include uncertainty, is high. Typically, application-specific software implementations are efficient enough. Their applicability, however, is limited to specific environments and specific goals, which is not the typical case for autonomous and intelligent systems. Dealing with complex environments that are not known a priori, and being able to accept wide-open goals requires these systems to reason and plan over generic domains, which are represented formally using logical languages. These requirements translate into algorithms that are computationally very challenging. Hence, there is a demanding need not only for new algorithms for domain-independent reasoning but also for novel hardware architectures that can accelerate these algorithms compared to software execution.

The Workshop on Accelerating Artificial Intelligence for Embedded Autonomy aims at gathering researchers and practitioners in the fields of autonomy, automated reasoning, planning algorithms, and embedded systems to discuss the development of novel hardware architectures that can accelerate the wide variety of AI algorithms demanded by advanced autonomous and intelligent systems. Topics of interest include hardware architectures and design methodologies to accelerate:

• Applications based on deep learning
• Skill-level and instinctive autonomy based on deep reinforcement learning
• Storage and retrieval of facts in knowledge bases
• Logical reasoning methods such as deduction
• Search for classical planning algorithms and Hierarchical Task Networks (HTN)
• Inference in probabilistic models such as Bayesian networks and probabilistic logic
• Planning algorithms for Markov Decision Processes (MDP) and Partial Observable Markov Decision Processes (POMDP)

The workshop is not limited to any particular hardware implementation platform. ASICs, FPGAs, and analog circuits, as well as heterogeneous platforms that combine hardware accelerators with CPUs and GPUs are all within the scope of the proposal. Similarly, emerging architectures such as those based on neuromorphic computing and so-called non Von Neumann machines are also considered in scope.

Submissions

The submission deadline is August 15, 2020.

Please submit your title, abstract, and a short bio by e-mail directly to pintoa@rtx.com.

Post-Workshop Publications

The workshop is focusing on the general issue of acceleration of AI techniques for embedded applications. We believe that, while a considerable amount of work has been done on neural networks, addressing the more general problem will require a much larger community of researchers to come together and contribute to the field. We are planning to create position papers in this area and perhaps a special issue on the subject where workshop presenters and participants, as well as the research community at large, can discuss their vision and current research work.