S.P. Arun, IISc Bangalore

Talk Title: If We Can Make Computers Play Chess, Why Can't We Make Them See?

Abstract: If we can make computers play chess and even Jeopardy, then why can't we make them see like us? This is a particularly perplexing question when we consider how we are still asked to recognize distorted letters on websites, which stand as daily proof that computers can't even recognize letters. What makes vision such a hard problem? How does the brain accomplish vision? To answer these questions we are recording brain signals at multiple scales to understand how the brain is solving this problem and using these data to improve computer vision. I will describe some of our recent work in which we have found striking similarities as well as differences between machine and human vision, and have used these differences to improve machine vision.

Saket Navlakha, Cold Spring Harbour Laboratory

Talk Title: Algorithms and data structures in the fruit fly brain

Abstract: A fundamental challenge in neuroscience is to understand the algorithms that neural circuits have evolved to solve computational problems critical for survival. In this talk, I will describe how the olfactory circuit in the fruit fly brain has evolved simple yet effective algorithms to process and store odors. First, I will describe how fruit flies use a variant of a traditional computer science algorithm (called locality-sensitive hashing) to perform efficient similarity searches. Second, I will describe how this circuit uses a variant of a classic data structure (called a Bloom filter) to perform novelty detection for odors. In both cases, we show that tricks from biology can be translated to improve machine computation, while also raising new hypotheses about neural function.

Tomaso Poggio, MIT

Talk Title: The Science and the Engineering of Intelligence

Abstract: In recent years, artificial intelligence researchers have built impressive systems. Two of my former postdocs — Demis Hassabis and Amnon Shashua — are behind two main recent success stories of AI: AlphaGo and Mobileye, based on two key algorithms, both originally suggested by discoveries in neuroscience: deep learning and reinforcement learning. But now recent engineering advances of the last 4 years — such as transformers, perceivers and MLP mixers— prompt an interesting question: will science or engineering win the race for AI? Do we need to understand the brain in order to build intelligent machines? or not? A deeper question is whether there exist a theoretical explanation — a common motif — for those network architectures, including the human brain, that perform so well in learning tasks. I will discuss the conjecture that for computable functions of many variables, compositional sparsity is equivalent to computability in polynomial time.

Subbarao Kambhampati, Arizona State university

Talk Title: Symbols as a Lingua Franca for Supporting Human-AI Interaction For Explainable and Advisable AI Systems

Abstract: Despite the surprising power of many modern AI systems that often learn their own representations, there is significant discontent about their inscrutability and the attendant problems in their ability to interact with humans. While alternatives such as neuro-symbolic approaches have been proposed, there is a lack of consensus on what they are about. There are often two independent motivations (i) symbols as a lingua franca for human-AI interaction and (ii) symbols as system-produced abstractions used by the AI system in its internal reasoning. The jury is still out on whether AI systems will need to use symbols in their internal reasoning to achieve general intelligence capabilities. Whatever the answer there is, the need for (human-understandable) symbols in human-AI interaction seems quite compelling. In particular, humans would be interested in providing explicit (symbolic) knowledge and advice -- and expect machine explanations in kind. This alone requires AI systems to maintain a symbolic interface for interaction with humans. In this talk, I will motivate this point of view, and describe recent efforts in our research group along this direction.

Jinjun Xiong, University of Buffalo

Talk Title: VisualNet: A Novel Statistical Distribution-based Deep Neural Network Model and its Connection to the Human Visual System

Abstract: The impressive results achieved by deep neural networks (DNNs) in various tasks, computer vision in particular, such as image recognition, object detection and image segmentation, have sparked the recent surging interests in artificial intelligence (AI) from both the industry and the academia alike. The wide adoption of DNN models in real-time applications has, however, brought up a need for more effective training of an easily parallelizable DNN model for low latency and high throughput. This is particularly challenging because of DNN's deep structures. To address this challenge, we observe that most of existing DNN models operate on deterministic numbers and process one single frame of image at a time, and may not fully utilize the temporal and contextual correlation typically present in multiple channels of the same image or adjacent frames from a video. Based on well-established statistical timing analysis foundations from the EDA domain, we propose a novel statistical distribution-based DNN model that extends existing DNN architectures but operates directly on correlated distributions rather than deterministic numbers. This new perspective of training DNN has resulted in surprising effects on achieving not only improved learning accuracy, but also reduced latency and increased high throughputs. Moreover, we show how such a network seems to be a a more realistic capture of the end-to-end human visual system. Preliminary experimental results on various tasks, including image classification, object detection, and 3D Cardiac Cine MRI segmentation, showed a great potential of this new type of DNN model (which we call VisualNet), which warrants further investigation.

Richa Singh, IIT Jodhpur

Talk Title: TBD

Abstract: TBD

V. Ramaswamy, BITS Pilani, Hyderabad

Talk Title: Understanding Brains & Deep Networks - Past, Present and Future

Abstract: A central goal of Neuroscience has been in building an understanding of the principles that govern computation in the brain. As a result, the field collectively has contemplated the question of what understanding might mean, what form it would take, and how we can get to it. Classically, Neuroscience was significantly limited in the availability of experimental tools, especially at the neural circuit level, which restricted our ability to operationalise these questions. More recent revolutionary advances in experimental techniques, however, have removed many of these barriers and we are confronted again with how exactly we should use these powerful techniques to distill an understanding of how brains work. Parallelly, over the last 10 years, the field of Deep Learning has leapfrogged AI into an era where multiple classical problems that were unsolved for decades are seeing successful everyday deployment. These advances have, however, not come with a concomitant deep understanding of the principles that govern computation in Deep Networks. Some argue that a lack of such understanding is hindering further advances, although such an understanding is also important for it’s own sake and in possibly inspiring hypotheses in Neuroscience. In this talk, I will survey some of these developments, in the past and present, and touch upon some of our own work in this direction. This is an extremely active and exciting area of research both in Neuroscience and in Machine Learning and indeed many of the important advances are arguably still ahead of us.

Dipanjan Roy, IIT Jodhpur

Talk Title: Bayesian Inference in Time and Disruption of the Prefrontal Code of Emotion Dynamics with Aging

Abstract: Affective experiences drive several higher-order cognitive processes, complex social behaviors, and guide perception, which is why its change with age has been an important area of research for years. Empirical evidence hints towards a bias associated with older individuals while experiencing positive emotions which has mainly been attributed to the ‘positivity effect’. While previous studies mostly used the two-dimensional framework of discrete emotions to investigate these changes, controlled stimuli ignored the complexity and temporal dynamics of real-world affective experiences. Crucially, owing to the lack of a theoretical or computational model, the ‘positivity effect’ has failed to explain opposing evidence in varied contexts. Here, we used naturalistic movie-watching tasks with fMRI to investigate whether the differential emotional experience of older individuals can be attributed to the change in the representation of uncertainty about future uncertainty estimation and outcomes. Our hypothesis is grounded on recent evidence showing that uncertainty is critical in modulating human emotion dynamics. The Bayesian Brain hypothesis posits that an accurate uncertainty estimation is crucial for optimal perception and higher-order cognitive processes. This, combined with the evidence from reinforcement learning shows an incomplete representation of uncertainty in older individuals propelled us to investigate our hypothesis from a representational and computational aspect, which remains critically underexplored in aging literature.

Dip Sankar Banerjee, IIT Jodhpur

Talk Title: Brain Inspired Computing and Systems

Abstract: Recent advances in VLSI and systems design has led to significant success in the general area of machine learning, specifically Deep Learning. However with the limitations imposed by Dennard Scaling, and the power wall, newer investigations have been necessitated towards future systems that can move away from von-Neumann philosophies keeping intact the usabilitties of general purpose processors. Brain inspired designs are now considered to be the "holy-grail" of future breakthroughs given the massive computations that the human brain can perform at a minimal power requirement. It has led to several innovations and designs that are slowly paving the path towards the future of computing. In this talk, we shall see an overview of these designs, the fundamental research that is happening and some of the special architectures along these lines that have found significant scientific as well as commercial success. We shall also discuss some of the future directions where we can look to innovate in an interdisciplinary manner.

Rohan Paul, IIT Delhi

Talk Title: Bridging the Semantic Gap Between Humans and Robots

Abstract: Robots are transitioning from restricted isolates environments to domains where they will increasingly interact with humans. Applications include robots in the factories, homes, search and rescue etc. In order to work along side humans, robots must possess a cognitive “human-like” understanding of the world. This talk will explore an AI-based framework for allowing robots to follow high-level instructions from a human, plan and execute semantic tasks and physically explore the environment to resolve ambiguity.

Amit Bhardwaj, IIT Jodhpur

Talk Title: Data-driven Haptics: Applications in Medical simulators and Telemedicine

Abstract: Unlike other senses, the haptics- the sense of touch has not been explored much. Because of the advancement in the field of machine haptics and availability of high-quality haptic interfaces, haptics has gained a lot of attention in the field of medical training and teleoperation. In this talk, I will be talking about the applications of machine learning in haptics for developing medical simulators and tele-medicine solutions.