Saurav Jha

ORCID: 0000-0001-7967-0039
Publications
Citations
Views
---
Saved
---
About
Contact & Profiles
Research Areas
  • Topic Modeling
  • Domain Adaptation and Few-Shot Learning
  • Natural Language Processing Techniques
  • Multimodal Machine Learning Applications
  • Human Pose and Action Recognition
  • COVID-19 diagnosis using AI
  • Vascular anomalies and interventions
  • Face and Expression Recognition
  • Face recognition and analysis
  • Biometric Identification and Security
  • Context-Aware Activity Recognition Systems
  • Speech Recognition and Synthesis
  • Text Readability and Simplification
  • Geophysical Methods and Applications
  • Advanced Image and Video Retrieval Techniques
  • Advanced Neural Network Applications
  • Machine Learning and Data Classification
  • Abdominal vascular conditions and treatments
  • Hip and Femur Fractures
  • Anomaly Detection Techniques and Applications
  • Authorship Attribution and Profiling
  • Sparse and Compressive Sensing Techniques
  • Bioactive Natural Diterpenoids Research
  • Cardiac pacing and defibrillation studies
  • Handwritten Text Recognition Techniques

Manipal College of Medical Sciences
2023-2025

Patan Academy of Health Sciences
2025

Chitwan Medical College
2024

UNSW Sydney
2022-2023

Netaji Subhas University of Technology
2023

University of St Andrews
2021

Motilal Nehru National Institute of Technology
2017-2019

10.22541/au.174055686.65895220/v1 preprint Authorea (Authorea) 2025-02-26

ABSTRACT Early identification and multidisciplinary management of complex conditions such as COXPD‐38 are crucial for optimizing outcomes in pediatric patients. Ongoing monitoring metabolic status, developmental progress, nutritional needs is essential supporting growth improving quality life.

10.1002/ccr3.70273 article EN cc-by-nc Clinical Case Reports 2025-02-26

Continual learning is an emerging research challenge in human activity recognition (HAR). As increasing number of HAR applications are deployed real-world environments, it important and essential to extend the model adapt change people’s routine. Otherwise, can become obsolete fail deliver activity-aware services. The existing has focused on detecting abnormal sensor events or new activities, however, extending currently under-explored. To directly tackle this challenge, we build recent...

10.1145/3440036 article EN ACM Transactions on Internet of Things 2021-03-27

In this paper, we investigate the continual learning of Vision Transformers (ViT) for challenging exemplar-free scenario, with special focus on how to efficiently distill knowledge its crucial self-attention mechanism (SAM). Our work takes an initial step towards a surgical investigation SAM designing coherent methods in ViTs. We first carry out evaluation established regularization techniques. then examine effect when applied two key enablers SAM: (a) contextualized embedding layers, their...

10.1109/cvprw56347.2022.00427 article EN 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) 2022-06-01

Continual learning (CL) aims to help deep neural networks learn new knowledge while retaining what has been learned. Recently, pre-trained vision-language models such as CLIP, with powerful generalization ability, have gaining traction practical CL candidates. However, the domain mismatch between pre-training and downstream tasks calls for finetuning of CLIP on latter. The deterministic nature existing methods makes them overlook many possible interactions across modalities deems unsafe...

10.48550/arxiv.2403.19137 preprint EN arXiv (Cornell University) 2024-03-28

In this work, we propose a novel method (GLOV) enabling Large Language Models (LLMs) to act as implicit Optimizers for Vision-Langugage (VLMs) enhance downstream vision tasks. Our GLOV meta-prompts an LLM with the task description, querying it suitable VLM prompts (e.g., zero-shot classification CLIP). These are ranked according purity measure obtained through fitness function. each respective optimization step, fed in-context examples (with their accuracies) equip knowledge of type text...

10.48550/arxiv.2410.06154 preprint EN arXiv (Cornell University) 2024-10-08

Given the recent deep learning advancements in face detection and recognition techniques for human faces, this paper answers question "how well would they work cartoons'?" - a domain that remains largely unexplored until recently, mainly due to unavailability of large scale datasets failure traditional methods on these. Our studies extends multiple frameworks aforementioned tasks. For detection, we incorporate Multi-task Cascaded Convolutional Network (MTCNN) architecture contrast it with...

10.48550/arxiv.1804.01753 preprint EN cc-by arXiv (Cornell University) 2018-01-01

Given the growing trend of continual learning techniques for deep neural networks focusing on domain computer vision, there is a need to identify which these generalizes well other tasks such as human activity recognition (HAR). As recent methods have mostly been composed loss regularization terms and memory replay, we provide constituent-wise analysis some prominent task-incremental employing HAR datasets. We find that most approaches lack substantial effect an intuition when they fail....

10.48550/arxiv.2007.03032 preprint EN other-oa arXiv (Cornell University) 2020-01-01

Open-World Compositional Zero-Shot Learning (OW-CZSL) aims to recognize new compositions of seen attributes and objects. In OW-CZSL, methods built on the conventional closed-world setting degrade severely due unconstrained OW test space. While previous works alleviate issue by pruning according external knowledge or correlations in pairs, they introduce biases that harm generalization. Some thus predict state object with independently constructed trained classifiers, ignoring are highly...

10.1109/iccv51070.2023.00171 article EN 2021 IEEE/CVF International Conference on Computer Vision (ICCV) 2023-10-01

The standard approaches to neural network implementation yield powerful function approximation capabilities but are limited in their abilities learn meta representations and reason probabilistic uncertainties predictions. Gaussian processes, on the other hand, adopt Bayesian learning scheme estimate such constrained by efficiency capacity. Neural Processes Family (NPF) intends offer best of both worlds leveraging networks for meta-learning predictive uncertainties. Such potential has brought...

10.48550/arxiv.2209.00517 preprint EN cc-by arXiv (Cornell University) 2022-01-01

In this paper, we present two approaches to automatically classify the gender of blog authors: first is a manual feature extraction based system incorporating novel classes: variable length character sequence patterns and thirteen new word classes, along with an added class surface features while second first-ever application memory variant Recurrent Neural Networks, i.e. Bidirectional Long Short Term Memory Networks (BLSTMs) on task. We use data sets report our results: well-explored one...

10.1109/icoac.2017.8441506 article EN 2017-12-01

Renal cell carcinoma (RCC) accounts for ~2% of global cancers and deaths. Survival depends on initial staging shows poor survival rate in metastatic disease. CT MRI are used evaluating RCC, PET/CT is disease assessment. We report a case where both 18F-FDG 68Ga-PSMA showed increased uptake liver lesions; however, subhepatic peritoneal deposit only PSMA. Also, lesions were seen better PSMA owing to lesser background uptake, suggesting possibility being potential tracer RCC evaluation.

10.1097/rlu.0000000000004648 article EN Clinical Nuclear Medicine 2023-04-19

The ambiguities introduced by the recombination of morphemes constructing several possible inflections for a word makes prediction syntactic traits in Morphologically Rich Languages (MRLs) notoriously complicated task. We propose Multi Task Deep Morphological analyzer (MT-DMA), character-level neural morphological based on multitask learning word-level tag markers Hindi and Urdu. MT-DMA predicts set six tags words Indo-Aryan languages: Parts-of-speech (POS), Gender (G), Number (N), Person...

10.48550/arxiv.1811.08619 preprint EN cc-by arXiv (Cornell University) 2018-01-01

While neural approaches using deep learning are the state-of-the-art for natural language processing (NLP) today, pre-neural algorithms and still find a place in NLP textbooks courses of recent years. In this paper, we compare two introductory taught Australia India, examine how Transformer balanced within lecture plan assessments courses. We also draw parallels with objects-first objects-later debate CS1 education. observe that add value to student by building an intuitive understanding...

10.48550/arxiv.2405.09854 preprint EN arXiv (Cornell University) 2024-05-16

Personalized text-to-image diffusion models have grown popular for their ability to efficiently acquire a new concept from user-defined text descriptions and few images. However, in the real world, user may wish personalize model on multiple concepts but one at time, with no access data previous due storage/privacy concerns. When faced this continual learning (CL) setup, most personalization methods fail find balance between acquiring retaining ones -- challenge that (CP) aims solve....

10.48550/arxiv.2410.00700 preprint EN arXiv (Cornell University) 2024-10-01

ABSTRACT Early recognition and prompt intervention are crucial in managing aconite poisoning. Rapid treatment with intravenous magnesium sulfate amiodarone can stabilize severe cardiac arrhythmias. Vigilant monitoring tailored therapeutic strategies enhance recovery improve patient outcomes acute poisoning cases.

10.1002/ccr3.9719 article EN cc-by-nc Clinical Case Reports 2024-12-01

ABSTRACT Timely diagnosis and surgical intervention are crucial in managing hydrocele of the canal Nuck. A systematic approach, including thorough examination appropriate imaging, followed by meticulous technique, ensures successful treatment favorable long‐term outcomes for patients.

10.1002/ccr3.70003 article EN cc-by-nc Clinical Case Reports 2024-12-01

Open-World Compositional Zero-Shot Learning (OW-CZSL) aims to recognize new compositions of seen attributes and objects. In OW-CZSL, methods built on the conventional closed-world setting degrade severely due unconstrained OW test space. While previous works alleviate issue by pruning according external knowledge or correlations in pairs, they introduce biases that harm generalization. Some thus predict state object with independently constructed trained classifiers, ignoring are highly...

10.48550/arxiv.2303.00404 preprint EN other-oa arXiv (Cornell University) 2023-01-01
Coming Soon ...