Day - 3
30 July 2022
9 AM – 9:45 AM
In this talk Dr. Arpan will introduce Device Edge Computing as one of the means for creating Intelligent Embedded cyber-physical Systems leading towards AI-driven IoT (AIoT). He will discuss some application use-cases of the same in different business domains like Manufacturing, Retail, Healthcare, Oil & Gas, SpaceTech etc. The talk will also introduce some tools and techniques like Signal Processing driven AI, Network Architecture Search, Auto-pruning etc. to convert large-AI models suitable for cloud to Tiny-AI models suitable for frugal, resource constrained edge devices. Finally Dr. Arpan will conclude with introduction to some futuristic technologies in this space, like Neuromorphic Computing, Metamaterials and Nano-sensing Systems.
9:45 AM – 10:30 AM
Computing is undergoing a paradigm shift: treating programs as mechanical procedures that realize a mathematical specification is slowly being replaced by a host of statistical learning methods that exploit the power of data abundance, preternatural generalization abilities, and efficient hardware platforms. While this is perhaps an acceptable approach for programs that decide what new product to suggest to a potential customer, it is potentially a recipe for disasters when applied to the paradigm of safety-critical systems. The question this talk seeks to answer is: why can we not have the best of both worlds? Can we combine learning techniques (for example reinforcement learning (RL) and deep RL) with traditional techniques for the design of safety-critical systems? We will look at some recent work where we are able to perform reward shaping from high-level temporal logic specifications such that any off-the-shelf RL algorithm that maximizes the expected long-term returns is guaranteed to maximize the probability of satisfying a given temporal logic specification. This result lets us not worry about good hand-crafted rewards that guarantee desired and safe behaviour, but instead uses mathematically precise high-level specifications to automatically infer such rewards. We will also examine how logical specifications can be leveraged to drastically reduce the number of demonstrations required to learn control policies from user demonstrations. Again, the objective is to move away from inferring implied task objectives (and control policies) from expert demonstrations to using temporal logic to formally specify high-level task objectives and couple it with non-expert demonstrations to infer optimal control policies. The main agenda of this talk is to argue for a logic-based framework to provide formal safety arguments for autonomous systems without sacrificing the advances provided by learning.
10:30 AM – 11 AM
[Tea Break + TCS Research Cafe]
11 AM – 11:45 AM
Optimal transport or OT has become a relevant tool in various machine learning applications. The rising popularity of OT is because it gives a principled approach in exploiting the underlying metric geometry to develop different notions of distances between probability distributions to be used in downstream applications. Consequently, there have been many works on developing computationally efficient tools to solve OT problems.
In this presentation, we look at the Riemannian approach to solving OT related optimization problems. The Riemannian approach has also become popular for solving structured nonlinear optimization problems. To that end, we discuss some of the recent works on exploiting Riemannian manifold structures to develop optimization-related ingredients for tackling OT formulations. We also look at various challenges and opportunities that lay ahead in making Riemannian tools a viable alternative for OT practitioners.
[TCS Research Cafe]
11:45 AM – 12:30 PM
Recent developments in computational intelligence, machine learning and soft computing have greatly expanded the horizon of data-driven modelling. However, time series modelling in general and chaotic time series modelling, in particular are still challenging problems of this domain. Soft Computing, which encompasses a plethora of model free approaches ranging from the well- established fuzzy logic to the emerging self-evolving neural networks, has established its credentials in this field of modelling. An attempt will be made in this session to showcase the capabilities of soft computing based models for modelling of chaotic time series and prediction of its future values.
As the name suggests, chaotic time series modelling is based on analysing the data from a system for finding connections between the past values of system state variables (input, internal and output variables) without explicit knowledge of the uncertainties and anomalous phenomena governing the physical behaviour of complex systems. Moreover, such systems often exhibit extreme sensitivity to initial conditions.
Neuro-fuzzy models are capable of self-learning the models of such systems from observed past values of data and identify their underlying interconnections, despite the uncertainty and irregular periodicity deeply ingrained and embedded in the time series data obtained from such systems. These models can be optimised by applying genetic algorithms. Hence, soft computing based hybrid models have been developed by our research group and applied for benchmark and application time series modelling and prediction. Some examples of successful applications will be presented in this session.
12:30 PM – 2 PM
2 PM – 3:30 PM
|Realizing Linear Controllers for Quadruped Robots on Planetary Terrains
|Barrier Coverage in Image-based Sensor Networks in the Presence of Occluding Objects
|Veermata Jijabai Institute of Technology (VJTI), Mumbai
|Hybrid Optimization Enabled Robust CNN-LSTM Technique for Network Intrusion Detection
|Pankaj Kumar Mishra
|On Controller Design for Unknown Nonlinear Systems with Prescribed Performance and Input Constraints
|Generating Motion Flow Maps with Attention in Time and RoAM, a new Robotic Dataset
|Fusion-UWnet: Multi-channel Fusion-based Deep CNN for Underwater Image Enhancement