no code implementations • 20 Mar 2024 • Zeyu Liu, Souvik Kundu, Anni Li, Junrui Wan, Lianghao Jiang, Peter Anthony Beerel
While compared in terms of runtime, AFLoRA can yield up to $1. 86\times$ improvement as opposed to similar PEFT alternatives.
1 code implementation • 8 Mar 2024 • Hao Kang, Qingru Zhang, Souvik Kundu, Geonhwa Jeong, Zaoxing Liu, Tushar Krishna, Tuo Zhao
Key-value (KV) caching has become the de-facto to accelerate generation speed for large language models (LLMs) inference.
no code implementations • 19 Feb 2024 • Souvik Kundu, Anthony Sarah, Vinay Joshi, Om J Omer, Sreenivas Subramoney
With the recent growth in demand for large-scale deep neural networks, compute in-memory (CiM) has come up as a prominent solution to alleviate bandwidth and on-chip interconnect bottlenecks that constrain Von-Neuman architectures.
no code implementations • 8 Feb 2024 • Sreetama Sarkar, Souvik Kundu, Peter A. Beerel
Our experimental evaluations show that RLNet can yield models with up to 11. 14x fewer ReLUs, with accuracy close to the all-ReLU models, on clean, naturally perturbed, and gradient-based perturbed images.
no code implementations • 11 Dec 2023 • Subhajit Dutta Chowdhury, Zhiyu Ni, Qingyuan Peng, Souvik Kundu, Pierluigi Nuzzo
By iteratively applying ARGS to prune both the perturbed graph adjacency matrix and the GNN model weights, we can find adversarially robust graph lottery tickets that are highly sparse yet achieve competitive performance under different untargeted training-time structure attacks.
no code implementations • 7 Dec 2023 • Yuhang Li, Youngeun Kim, DongHyun Lee, Souvik Kundu, Priyadarshini Panda
In the realm of deep neural network deployment, low-bit quantization presents a promising avenue for enhancing computational efficiency.
no code implementations • 2 Dec 2023 • Souvik Kundu, Rui-Jie Zhu, Akhilesh Jaiswal, Peter A. Beerel
Neuromorphic computing and, in particular, spiking neural networks (SNNs) have become an attractive alternative to deep neural networks for a broad range of signal processing applications, processing static and/or temporal inputs from different sensory modalities, including audio and vision sensors.
no code implementations • 2 Oct 2023 • Hongyi Wang, Felipe Maia Polo, Yuekai Sun, Souvik Kundu, Eric Xing, Mikhail Yurochkin
Training AI models that generalize across tasks and domains has long been among the open problems driving AI research.
1 code implementation • 29 Sep 2023 • Lu Yin, Ajay Jaiswal, Shiwei Liu, Souvik Kundu, Zhangyang Wang
Contrary to this belief, this paper presents a counter-argument: small-magnitude weights of pre-trained model weights encode vital knowledge essential for tackling difficult downstream tasks - manifested as the monotonic relationship between the performance drop of downstream tasks across the difficulty spectrum, as we prune more pre-trained weights by magnitude.
no code implementations • 29 Aug 2023 • Sharath Nittur Sridhar, Souvik Kundu, Sairam Sundaresan, Maciej Szankin, Anthony Sarah
However, training super-networks from scratch can be extremely time consuming and compute intensive especially for large models that rely on a two-stage training process of pre-training and fine-tuning.
1 code implementation • 6 Aug 2023 • Yue Hu, Xinan Ye, Yifei Liu, Souvik Kundu, Gourav Datta, Srikar Mutnuri, Namo Asavisanu, Nora Ayanian, Konstantinos Psounis, Peter Beerel
This paper presents "FireFly", a synthetic dataset for ember detection created using Unreal Engine 4 (UE4), designed to overcome the current lack of ember-specific training resources.
no code implementations • 14 Jul 2023 • Souvik Kundu, Sharath Nittur Sridhar, Maciej Szankin, Sairam Sundaresan
In this paper, we present Sensi-BERT, a sensitivity driven efficient fine-tuning of BERT models that can take an off-the-shelf pre-trained BERT model and yield highly parameter-efficient models for downstream tasks.
1 code implementation • 10 Jun 2023 • Yonggan Fu, Ye Yuan, Souvik Kundu, Shang Wu, Shunyao Zhang, Yingyan Lin
Generalizable Neural Radiance Fields (GNeRF) are one of the most promising real-world solutions for novel view synthesis, thanks to their cross-scene generalization capability and thus the possibility of instant rendering on new scenes.
no code implementations • 26 Apr 2023 • Souvik Kundu, Yuke Zhang, Dake Chen, Peter A. Beerel
Large number of ReLU and MAC operations of Deep neural networks make them ill-suited for latency and compute-efficient private inference.
no code implementations • 6 Apr 2023 • Md Abdullah-Al Kaiser, Gourav Datta, Sreetama Sarkar, Souvik Kundu, Zihan Yin, Manas Garg, Ajey P. Jacob, Peter A. Beerel, Akhilesh R. Jaiswal
The massive amounts of data generated by camera sensors motivate data processing inside pixel arrays, i. e., at the extreme-edge.
no code implementations • 17 Feb 2023 • Shashank Nag, Gourav Datta, Souvik Kundu, Nitin Chandrachoodan, Peter A. Beerel
Vision Transformer models, such as ViT, Swin Transformer, and Transformer-in-Transformer, have recently gained significant traction in computer vision tasks due to their ability to capture the global relation between features which leads to superior performance.
no code implementations • 23 Jan 2023 • Souvik Kundu, Shunlin Lu, Yuke Zhang, Jacqueline Liu, Peter A. Beerel
For a similar ReLU budget SENet can yield models with ~2. 32% improved classification accuracy, evaluated on CIFAR-100.
1 code implementation • ICCV 2023 • Yan Han, Peihao Wang, Souvik Kundu, Ying Ding, Zhangyang Wang
In this paper, we enhance ViG by transcending conventional "pairwise" linkages and harnessing the power of the hypergraph to encapsulate image information.
no code implementations • ICCV 2023 • Yuke Zhang, Dake Chen, Souvik Kundu, Chenghao Li, Peter A. Beerel
Then, given our observation that external attention (EA) presents lower PI latency than widely-adopted self-attention (SA) at the cost of accuracy, we present a selective attention search (SAS) method to integrate the strength of EA and SA.
no code implementations • 27 Dec 2022 • Souvik Kundu, Sairam Sundaresan, Sharath Nittur Sridhar, Shunlin Lu, Han Tang, Peter A. Beerel
Existing deep neural networks (DNNs) that achieve state-of-the-art (SOTA) performance on both clean and adversarially-perturbed images rely on either activation or weight conditioned convolution operations.
no code implementations • 21 Dec 2022 • Gourav Datta, Zeyu Liu, Md Abdullah-Al Kaiser, Souvik Kundu, Joe Mathai, Zihan Yin, Ajey P. Jacob, Akhilesh R. Jaiswal, Peter A. Beerel
Although the overhead for the first layer MACs with direct encoding is negligible for deep SNNs and the CV processing is efficient using SNNs, the data transfer between the image sensors and the downstream processing costs significant bandwidth and may dominate the total energy.
no code implementations • 16 Sep 2022 • Fang Chen, Gourav Datta, Souvik Kundu, Peter Beerel
With the aggressive down-sampling of the activation maps in the initial layers (providing up to 22x reduction in memory consumption), our approach achieves 1. 43% higher test accuracy compared to SOTA techniques with iso-memory footprints.
no code implementations • 29 Aug 2022 • Soumyabrata Talukder, Souvik Kundu, Ratnesh Kumar
Most sensor calibrations rely on the linearity and steadiness of their response characteristics, but practical sensors are nonlinear, and their response drifts with time, restricting their choices for adoption.
1 code implementation • 28 Aug 2022 • Yue Niu, Saurav Prakash, Souvik Kundu, Sunwoo Lee, Salman Avestimehr
However, the heterogeneous-client setting requires some clients to train full model, which is not aligned with the resource-constrained setting; while the latter ones break privacy promises in FL when sharing intermediate representations or labels with the server.
1 code implementation • 27 Aug 2022 • Sara Babakniya, Souvik Kundu, Saurav Prakash, Yue Niu, Salman Avestimehr
A possible solution to this problem is to utilize off-the-shelf sparse learning algorithms at the clients to meet their resource budget.
no code implementations • 7 Aug 2022 • Pavan Kumar Reddy Boppidi, Victor Jeffry Louis, Arvind Subramaniam, Rajesh K. Tripathy, Souri Banerjee, Souvik Kundu
The experimental results demonstrate that the proposed approach is very effective to separate image sources, and also the contrast of the images are improved with an improvement factor in terms of percentage of structural similarity as 67. 27% when compared with the software-based implementation of conventional ACY ICA and Fast ICA algorithms.
no code implementations • 28 May 2022 • Gourav Datta, Souvik Kundu, Zihan Yin, Joe Mathai, Zeyu Liu, Zixu Wang, Mulin Tian, Shunlin Lu, Ravi T. Lakkireddy, Andrew Schmidt, Wael Abd-Almageed, Ajey P. Jacob, Akhilesh R. Jaiswal, Peter A. Beerel
The designs also reduce the sensor and total energy (obtained from in-house circuit simulations at Globalfoundries 22nm technology node) per frame by 5. 7x and 1. 14x, respectively.
1 code implementation • LREC 2022 • Vladimir Araujo, Andrés Carvallo, Souvik Kundu, José Cañete, Marcelo Mendoza, Robert E. Mercer, Felipe Bravo-Marquez, Marie-Francine Moens, Alvaro Soto
Due to the success of pre-trained language models, versions of languages other than English have been released in recent years.
no code implementations • 28 Mar 2022 • Souvik Kundu, Sairam Sundaresan, Massoud Pedram, Peter A. Beerel
In this paper, we present a fast learnable once-for-all adversarial training (FLOAT) algorithm, which instead of the existing FiLM-based conditioning, presents a unique weight conditioned learning that requires no additional layer, thereby incurring no significant increase in parameter count, training time, or network latency compared to standard adversarial training.
no code implementations • 7 Mar 2022 • Gourav Datta, Souvik Kundu, Zihan Yin, Ravi Teja Lakkireddy, Joe Mathai, Ajey Jacob, Peter A. Beerel, Akhilesh R. Jaiswal
Visual data in such cameras are usually captured in the form of analog voltages by a sensor pixel array, and then converted to the digital domain for subsequent AI processing using analog-to-digital converters (ADC).
no code implementations • 24 Dec 2021 • Souvik Kundu, Shikai Wang, Qirui Sun, Peter A. Beerel, Massoud Pedram
Compared to the baseline FP-32 models, BMPQ can yield models that have 15. 4x fewer parameter bits with a negligible drop in accuracy.
no code implementations • NeurIPS 2021 • Souvik Kundu, Qirui Sun, Yao Fu, Massoud Pedram, Peter Beerel
Knowledge distillation (KD) has recently been identified as a method that can unintentionally leak private information regarding the details of a teacher model to an unauthorized student.
no code implementations • 28 Oct 2021 • Yang Hu, Connor Imes, Xuanang Zhao, Souvik Kundu, Peter A. Beerel, Stephen P. Crago, John Paul N. Walters
We propose EdgePipe, a distributed framework for edge systems that uses pipeline parallelism to both speed up inference and enable running larger (and more accurate) models that otherwise cannot fit on single edge devices.
no code implementations • 13 Oct 2021 • Digbalay Bose, Krishna Somandepalli, Souvik Kundu, Rimita Lahiri, Jonathan Gratch, Shrikanth Narayanan
Computational modeling of the emotions evoked by art in humans is a challenging problem because of the subjective and nuanced nature of art and affective signals.
1 code implementation • ICCV 2021 • Souvik Kundu, Massoud Pedram, Peter A. Beerel
Low-latency deep spiking neural networks (SNNs) have become a promising alternative to conventional artificial neural networks (ANNs) because of their potential for increased energy efficiency on event-driven neuromorphic hardware.
no code implementations • 29 Sep 2021 • Souvik Kundu, Peter Anthony Beerel, Sairam Sundaresan
In this paper, we present Fast Learnable Once-for-all Adversarial Training (FLOAT) which transforms the weight tensors without using extra layers, thereby incurring no significant increase in parameter count, training time, or network latency compared to a standard adversarial training.
no code implementations • 26 Jul 2021 • Gourav Datta, Souvik Kundu, Peter A. Beerel
This paper presents a training framework for low-latency energy-efficient SNNs that uses a hybrid encoding scheme at the input layer in which the analog pixel values of an image are directly applied during the first timestep and a novel variant of spike temporal coding is used during subsequent timesteps.
no code implementations • 26 Jul 2021 • Gourav Datta, Souvik Kundu, Akhilesh R. Jaiswal, Peter A. Beerel
However, the accurate processing of the spectral and spatial correlation between the bands requires the use of energy-expensive 3-D Convolutional Neural Networks (CNNs).
Computational Efficiency Hyperspectral Image Classification +1
no code implementations • 16 Jul 2021 • Souvik Kundu, Gourav Datta, Massoud Pedram, Peter A. Beerel
To evaluate the merits of our approach, we performed experiments with variants of VGG and ResNet, on both CIFAR-10 and CIFAR-100, and VGG16 on Tiny-ImageNet. The SNN models generated through the proposed technique yield SOTA compression ratios of up to 33. 4x with no significant drops in accuracy compared to baseline unpruned counterparts.
no code implementations • 21 Dec 2020 • Souvik Kundu, Sairam Sundaresan
We propose a novel framework for producing a class of parameter and compute efficient models called AttentionLitesuitable for resource-constrained applications.
no code implementations • 17 Dec 2020 • Souvik Kundu, Hesham Mostafa, Sharath Nittur Sridhar, Sairam Sundaresan
Convolutional layers are an integral part of many deep neural network solutions in computer vision.
1 code implementation • COLING 2020 • Qian Lin, Souvik Kundu, Hwee Tou Ng
One of the major challenges is that a dialogue system may generate an undesired utterance leading to a dialogue breakdown, which degrades the overall interaction quality.
1 code implementation • 3 Nov 2020 • Souvik Kundu, Mahdi Nazemi, Peter A. Beerel, Massoud Pedram
This paper presents a dynamic network rewiring (DNR) method to generate pruned deep neural network (DNN) models that are robust against adversarial attacks yet maintain high accuracy on clean images.
2 code implementations • 11 Aug 2020 • Qin Lyu, Kaushik Chakrabarti, Shobhit Hathi, Souvik Kundu, Jianwen Zhang, Zheng Chen
In this paper, we study how to leverage pre-trained language models in Text-to-SQL.
no code implementations • ACL 2020 • Souvik Kundu, Qian Lin, Hwee Tou Ng
Despite recent progress in conversational question answering, most prior work does not focus on follow-up questions.
1 code implementation • 29 Jan 2020 • Souvik Kundu, Mahdi Nazemi, Massoud Pedram, Keith M. Chugg, Peter A. Beerel
We also compared the performance of our proposed architectures with that of ShuffleNet andMobileNetV2.
no code implementations • 2 Oct 2019 • Souvik Kundu, Saurav Prakash, Haleh Akrami, Peter A. Beerel, Keith M. Chugg
To explore the potential of this approach, we have experimented with two widely accepted datasets, CIFAR-10 and Tiny ImageNet, in sparse variants of both the ResNet18 and VGG16 architectures.
1 code implementation • ACL 2019 • Souvik Kundu, Tushar Khot, Ashish Sabharwal, Peter Clark
To capture additional context, PathNet also composes the passage representations along each path to compute a passage-based representation.
1 code implementation • EMNLP 2018 • Souvik Kundu, Hwee Tou Ng
However, current approaches suffer from an impractical assumption that every question has a valid answer in the associated passage.
1 code implementation • 31 May 2018 • Sourya Dey, Diandian Chen, Zongyang Li, Souvik Kundu, Kuan-Wen Huang, Keith M. Chugg, Peter A. Beerel
We demonstrate an FPGA implementation of a parallel and reconfigurable architecture for sparse neural networks, capable of on-chip training and inference.
1 code implementation • 25 Jan 2018 • Souvik Kundu, Hwee Tou Ng
Neural network models recently proposed for question answering (QA) primarily focus on capturing the passage-question relation.
Ranked #4 on Question Answering on NewsQA