Author name cluster

Deepak Singh

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

5 papers

2 author rows

EAAI Journal 2025 Journal Article

A systematic review on vision-based gaze estimation: Advance in computer vision and deep learning

Sapna Singh Kshatri
Deepak Singh

Eye gaze estimation is all about figuring out where a person is looking by analyzing eye movements. This is usually done using cameras or eye-tracking devices, a key technique in computer vision. Gaze estimation has a wide range of uses—from helping people interact with computers more naturally (HCI), monitoring drivers, enhancing virtual reality experiences, and reading emotional cues in affective computing. Deep learning has improved how accurately and reliably these systems work in recent years. Models like Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Generative Adversarial Networks (GANs) have made it easier to detect and interpret eye movements in complex settings. This review closely examines how gaze estimation works, the datasets commonly used, and both traditional and deep learning-based methods. It also points out what's working well, where current methods fall short, and what researchers still need to solve. The goal is to offer helpful insights for anyone working to improve or apply gaze tracking technologies in real-world scenarios.

Details DOI

EAAI Journal 2023 Journal Article

AgrIntel: Spatio-temporal profiling of nationwide plant-protection problems using helpline data

Samarth Godara
Durga Toshniwal
Ram Swaroop Bana
Deepak Singh
Jatin Bedi
Rajender Parsad
Jai Prakash Singh Dabas
Abimanyu Jhajhria

Sustainable development of the national food system must ensure the introduction of adequate food security interventions and policies. However, several high-end technological developments remain unexplored, which can be used to gain explicit information regarding agricultural problems. In this direction, the presented work proposes AgrIntel, a framework consisting of multiple AI-based pipelines to process nationwide farmers’ helpline data and obtain spatiotemporal insights regarding food-production problems on an extensive scale. AgrIntel overcomes several limitations of the existing methods used for similar objectives, including limited scalability, low frequency, and high cost. The call-logs dataset used in the study is obtained from the nationwide network of farmers’ helpline centers, managed by the Ministry of Agriculture & Farmers’ Welfare, Government of India. The article demonstrates the Spatio-temporal profile of one of India’s highest food grain-affecting diseases, i. e. , “blast in rice crop”, to demonstrate the utility of the AgrIntel pipelines. First, the proposed framework extracts and clusters the precise geographical locations of farmers calling for help corresponding to the target agricultural problem. Next, the temporal modeling of the problem helps extract the critical dates corresponding to the crop disease/pest spread. Furthermore, by incorporating the historical agroclimatological data, the article introduces a new medium to extract the favorable weather conditions corresponding to the targeted disease/pest outbreak. In addition, the study explores the potential of Deep Learning models (based on Artificial Neural Network, Convolutional Neural Network, Gated Recurrent Unit and Long short-term memory unit) to efficiently predict the futuristic demand for assistance regarding target problems (RMSE of ≈ 1. 5 and MAE of ≈ 0. 9 query calls). The obtained results expose unrevealed insights regarding food production problems, significantly boosting the food security policy-designing procedure.

Details DOI

ICLR Conference 2023 Conference Paper

Sparse Distributed Memory is a Continual Learner

Trenton Bricken
Xander Davies
Deepak Singh
Dmitry Krotov
Gabriel Kreiman

Continual learning is a problem for artificial neural networks that their biological counterparts are adept at solving. Building on work using Sparse Distributed Memory (SDM) to connect a core neural circuit with the powerful Transformer model, we create a modified Multi-Layered Perceptron (MLP) that is a strong continual learner. We find that every component of our MLP variant translated from biology is necessary for continual learning. Our solution is also free from any memory replay or task information, and introduces novel methods to train sparse networks that may be broadly applicable.

Details

ICRA Conference 2022 Conference Paper

CCO-VOXEL: Chance Constrained Optimization over Uncertain Voxel-Grid Representation for Safe Trajectory Planning

Sudarshan S. Harithas
Rishabh Dev Yadav
Deepak Singh
Arun Kumar Singh 0001
K. Madhava Krishna

We present CCO-VOXEL: the very first chance-constrained optimization (CCO) algorithm that can compute trajectory plans with probabilistic safety guarantees in real-time directly on the voxel-grid representation of the world. CCO-VOXEL maps the distribution over the distance to the closest obstacle to a distribution over collision-constraint violation and computes an optimal trajectory that minimizes the violation probability. Importantly, unlike existing works, we never assume the nature of the sensor uncertainty or the probability distribution of the resulting collision-constraint violations. We leverage the notion of Hilbert Space embedding of distributions and Maximum Mean Discrepancy (MMD) to compute a tractable surrogate for the original chance-constrained optimization problem and employ a combination of A* based graph-search and Cross-Entropy Method for obtaining its minimum. We show tangible performance gain in terms of collision avoidance and trajectory smoothness as a consequence of our probabilistic formulation vis a vis state-of-the-art planning methods that do not account for such non-parametric noise. Finally, we also show how a combination of low-dimensional feature embedding and pre-caching of Kernel Matrices of MMD allow us to achieve real-time performance in simulations as well as in implementations on on-board commodity hardware that controls the quadrotor flight.

Details

AIIM Journal 2019 Journal Article

Compositional model based on factorial evolution for realizing multi-task learning in bacterial virulent protein prediction

Deepak Singh
Pradeep Singh
Dilip Singh Sisodia

The ability of multitask learning promulgated its sovereignty in the machine learning field with the diversified application including but not limited to bioinformatics and pattern recognition. Bioinformatics provides a wide range of applications for Multitask Learning (MTL) methods. Identification of Bacterial virulent protein is one such application that helps in understanding the virulence mechanism for the design of drug and vaccine. However, the limiting factor in a reliable prediction model is the scarcity of the experimentally verified training data. To deal with, casting the problem in a Multitask Learning scenario, could be beneficial. Reusability of auxiliary data from related multiple domains in the prediction of target domain with limited labeled data is the primary objective of multitask learning model. Due to the amalgamation of multiple related data, it is possible that the probability distribution between the features tends to vary. Therefore, to deal with change amongst the feature distribution, this paper proposes a composite model for multitask learning framework which is based on two principles: discovering the shared parameters for identifying the relationships between tasks and common underlying representation of features amongst the related tasks. Through multi-kernel and factorial evolution, the proposed framework able to discover the shared kernel parameters and latent feature representation that is common amongst the tasks. To examine the benefits of the proposed model, an extensive experiment is performed on the freely available dataset at VirulentPred web server. Based on the results, we found that multitask learning model performs better than the conventional single task model. Additionally, our findings state that if the distribution between the tasks is high, then training the multiple models yield slightly better prediction. However, if the data distribution difference is low, multitask learning significantly outperforms the individual learning.

Details DOI