Showing papers by "Saibal Mukhopadhyay published in 2023"

Heterogeneous Neuronal and Synaptic Dynamics for Spike-Efficient Unsupervised Learning: Theory and Design Principles

Abstract: This paper shows that the heterogeneity in neuronal and synaptic dynamics reduces the spiking activity of a Recurrent Spiking Neural Network (RSNN) while improving prediction performance, enabling spike-efficient (unsupervised) learning. We analytically show that the diversity in neurons' integration/relaxation dynamics improves an RSNN's ability to learn more distinct input patterns (higher memory capacity), leading to improved classification and prediction performance. We further prove that heterogeneous Spike-Timing-Dependent-Plasticity (STDP) dynamics of synapses reduce spiking activity but preserve memory capacity. The analytical results motivate Heterogeneous RSNN design using Bayesian optimization to determine heterogeneity in neurons and synapses to improve $\mathcal{E}$, defined as the ratio of spiking activity and memory capacity. The empirical results on time series classification and prediction tasks show that optimized HRSNN increases performance and reduces spiking activity compared to a homogeneous RSNN.

Roadmap for Unconventional Computing with Nanotechnology

Abstract: In the Beyond Moore Law era, with increasing edge intelligence, domain-specific computing embracing unconventional approaches will become increasingly prevalent. At the same time, the adoption of a wide variety of nanotechnologies will offer benefits in energy cost, computational speed, reduced footprint, cyber-resilience and processing prowess. The time is ripe to lay out a roadmap for unconventional computing with nanotechnologies to guide future research and this collection aims to fulfill that need. The authors provide a comprehensive roadmap for neuromorphic computing with electron spins, memristive devices, two-dimensional nanomaterials, nanomagnets and assorted dynamical systems. They also address other paradigms such as Ising machines, Bayesian inference engines, probabilistic computing with p-bits, processing in memory, quantum memories and algorithms, computing with skyrmions and spin waves, and brain inspired computing for incremental learning and solving problems in severely resource constrained environments. All of these approaches have advantages over conventional Boolean computing predicated on the von-Neumann architecture. With the computational need for artificial intelligence growing at a rate 50x faster than Moore law for electronics, more unconventional approaches to computing and signal processing will appear on the horizon and this roadmap will aid in identifying future needs and challenges.

Effect of prasugrel versus ticagrelor on coronary microcirculation in patients undergoing pharmacoinvasive strategy - acute and short-term results.

Abstract: INTRODUCTION Both ticagrelor and prasugrel are class I recommendations for treatment of ST-elevation myocardial infarction (STEMI) undergoing percutaneous coronary intervention (PCI) [1]. But clinical outcomes with the two drugs are conflicting which might be due to differential effects on coronary microcirculation. No study to date had compared the effects of prasugrel or ticagrelor on coronary microcirculation in patients undergoing pharmacoinvasive PCI (pPCI). AIM AND OBJECTIVE To compare the effects of prasugrel and ticagrelor on coronary microcirculation in STEMI patients undergoing pPCI as assessed by Myocardial Blush Grade (MBG). The secondary aim was to assess flow in the infarct-related artery by corrected thrombolysis in myocardial infarction (TIMI) frame count (cTFC) and whether a differential effect if detected on coronary microcirculation translated in improvement in left ventricular ejection fraction assessed at 6 months. MATERIAL AND METHODS A total of 240 patients with STEMI were evaluated in this open-label randomized control trial who initially underwent thrombolysis and later PCI (from 24 to 48 h) post-successful thrombolysis. The study subjects were randomized to receive either ticagrelor (n = 120) or prasugrel (n = 120) in 1 : 1 ratio 2 h prior to elective PCI. Patients underwent PCI according to standard protocol and post-procedure cTFC and MBG were compared. Patients were also followed up for 6 months to compare ejection fractions in both groups. We also assessed the effect of the two drugs on bleeding complications during hospitalization and over 6-month follow-up period. RESULTS There were no significant differences between the two groups with respect to baseline characteristics. Prasugrel administration resulted in higher MBG Grade 3 (50.86% vs 33.89%, P = 0.012) and lower cTFC (17.14 ± 4.08 vs 19.3 ± 4.06, P < 0.01). Improvement in ejection fraction was significantly higher with prasugrel compared to ticagrelor (10.29% ± 15.2 vs 4.66% ± 13.5, P = 0.003). Bleeding events at 6 months follow-up according to TIMI classification were similar in both the groups (11.86% vs 6.9%, P = 0.39). CONCLUSION Prasugrel produces greater improvement in coronary microcirculation than Ticagrelor resulting in improved myocardial salvage in patients of STEMI undergoing pPCI.

Unsupervised 3D Object Learning through Neuron Activity aware Plasticity

Abstract: We present an unsupervised deep learning model for 3D object classification. Conventional Hebbian learning, a well-known unsupervised model, suffers from loss of local features leading to reduced performance for tasks with complex geometric objects. We present a deep network with a novel Neuron Activity Aware (NeAW) Hebbian learning rule that dynamically switches the neurons to be governed by Hebbian learning or anti-Hebbian learning, depending on its activity. We analytically show that NeAW Hebbian learning relieves the bias in neuron activity, allowing more neurons to attend to the representation of the 3D objects. Empirical results show that the NeAW Hebbian learning outperforms other variants of Hebbian learning and shows higher accuracy over fully supervised models when training data is limited.

Brain-Inspired Spatiotemporal Processing Algorithms for Efficient Event-Based Perception

Abstract: Neuromorphic event-based cameras can unlock the true potential of bio-plausible sensing systems that mimic our human perception. However, efficient spatiotemporal processing algorithms must enable their low-power, low-latency, real-world application. In this talk, we highlight our recent efforts in this direction. Specifically, we talk about how brain-inspired algorithms such as spiking neural networks (SNNs) can approximate spatiotemporal sequences efficiently without requiring complex recurrent structures. Next, we discuss their event-driven formulation for training and inference that can achieve realtime throughput on existing commercial hardware. We also show how a brain-inspired recurrent SNN can be modeled to perform on event-camera data. Finally, we will talk about the potential application of associative memory structures to efficiently build representation for event-based perception.

Brain-Inspired Spiking Neural Network for Online Unsupervised Time Series Prediction

Abstract: Energy and data-efficient online time series prediction for predicting evolving dynamical systems are critical in several fields, especially edge AI applications that need to update continuously based on streaming data. However, current DNN-based supervised online learning models require a large amount of training data and cannot quickly adapt when the underlying system changes. Moreover, these models require continuous retraining with incoming data making them highly inefficient. To solve these issues, we present a novel Continuous Learning-based Unsupervised Recurrent Spiking Neural Network Model (CLURSNN), trained with spike timing dependent plasticity (STDP). CLURSNN makes online predictions by reconstructing the underlying dynamical system using Random Delay Embedding by measuring the membrane potential of neurons in the recurrent layer of the RSNN with the highest betweenness centrality. We also use topological data analysis to propose a novel methodology using the Wasserstein Distance between the persistence homologies of the predicted and observed time series as a loss function. We show that the proposed online time series prediction methodology outperforms state-of-the-art DNN models when predicting an evolving Lorenz63 dynamical system.

Learning Deep Neural Network Controller for Path Following of Unicycle Robots

Abstract: This letter investigates the scope of deep neural network (DNN) based controller in the path following task for unicycle mobile robots. A DNN-based controller is trained to follow paths with arbitrary curvature in two-dimensional space. The training process does not require initialization or supervision from any other known expert controller. Rather, the training of the DNN controller is guided by another predictive neural network that represents a path following error dynamics which is exponentially stable at the origin. The two DNNs are trained jointly in a simulated environment. The learned DNN controller is then employed as a standalone controller in a real unicycle robot for the tasks of following various linear and curved paths.

Psychosocial Stress and Well-being in patients presenting with Acute Myocardial Infarction in a Tertiary Care Center.

Abstract: Psychosocial factors such as stress have been previously implicated as a risk factor for cardiovascular diseases (CVDs). There is little evidence regarding the prevalence of stress among patients with acute myocardial infarction (AMI). A total of 903 patients with AMI enrolled in the North Indian ST-Segment Elevation Myocardial Infarction (NORIN-STEMI) registry were included in this study. Perceived stress in these subjects was evaluated using the Perceived Stress Scale-10 questionnaire while the World health Organization (WHO-5) Well-being Index was used to evaluate psychological well-being. All these patients were followed up for one month and major adverse cardiac events (MACE) were determined. A majority of patients with AMI had either severe (478 [52.9%]) or moderate stress (347 [38.4%]) while low stress levels were observed in 78 [8.6%] patients. Additionally, most of the patients with AMI (478 [53%]) had WHO-5 well-being index <50%. Subjects with severe stress were younger (50.86 ± 13.31; P < 0.0001), more likely to be males (403 [84.30%]; P = 0.027), were less likely to have optimal level of physical activity (P < 0.0001) and had lower WHO-5 well-being score (45.54 ± 1.94%; P < 0.0001) as compared to those with low and moderate stress levels. On 30-days follow-up, subjects with moderate/severe stress had higher MACE however, the difference was non-significant (2.1% vs 1.04%; P = 0.42). A high prevalence of perceived stress and low well-being index was observed in patients presenting with AMI in India.

A High Performance Computing Architecture for Real-Time Digital Emulation of RF Interactions

Abstract: A high performance architecture for emulating realtime radio frequency systems is presented. The architecture is developed based on a novel compute model and uses nearmemory techniques coupled with highly distributed autonomous control to simultaneously optimize throughput and minimize latency. A cycle level C++ based simulator is used to validate the proposed architecture with simulation of complex RF scenarios.

29.1 A 32.5mW Mixed-Signal Processing-in-Memory-Based k-SAT Solver in 65nm CMOS with 74.0% Solvability for 3D-Variable 126-Clause 3-SAT Problems

Abstract: Boolean satisfiability (k-SAT, k ≥3) is an NP-complete combinatorial optimization problem (COP) with applications in communication, flight network, supply chain and finance, to name a few. The ASICs for SAT and other COP solvers have been demonstrated using continuous-time dynamics [1], simulated annealing [2], oscillator interaction [3] and stochastic automata annealing [4]. However, prior designs show low solvability for complex problems ([1] shows 16% solvability for 30 variables and 126 clauses), and use a small, fixed network topology (King's graph [3] or Lattice Graph [2] or 3-SAT [1]) limiting the flexibility of problem solving. A digital fully connected processor enables flexibility but incurs a large area, latency and power overhead [4]. This paper presents a k-SAT solver where a Continuous-Time Stochastic Recurrent Neural Network (CT-SRNN), controlled by a Discrete-Time Finite-State-Machine (DT-FSM), uses unsupervised learning to search for an optimal solution (Fig. 29.1.1). A 65nm test-chip based on a Mixed-Signal Processing-in-Memory (MS-PIM) architecture is presented. Measured results demonstrate a higher solvability (74.0% for 30 variables and 126 clauses, vs. 16% in [1]) and an improved flexibility (k > 3, different number of variables per clause) in mapping k-SAT problems.

A ssociative m emory augmented a syn - chronous spatiotemporal r epresentation learning for e vent - based perception

Abstract: We propose EventFormer – a computationally efficient event-based representation learning framework for asynchronously processing event camera data. EventFormer treats sparse input events as a spatially unordered set and models their spatial interactions using self-attention mechanism. An associative memory-augmented recurrent module is used to correlate with the stored representation computed from past events. A memory addressing mechanism is proposed to store and retrieve the latent states only where these events occur and update them only when they occur. The representation learning shift from input space to the latent memory space resulting in reduced computation cost for processing each event. We show that EventFormer achieves 0 . 5% and 9% better accuracy with 30000 × and 200 × less computation compared to the state-of-the-art dense and event-based method, respectively, on event-based object recognition datasets.