A5024017460- Domain Adaptation and Few-Shot Learning
- Handwritten Text Recognition Techniques
- Imbalanced Data Classification Techniques
- 3D Surveying and Cultural Heritage
- Human Pose and Action Recognition
- Image Processing and 3D Reconstruction
- Visual Attention and Saliency Detection
- Advanced Image and Video Retrieval Techniques
- Image Retrieval and Classification Techniques
- Machine Learning and Algorithms
- Hate Speech and Cyberbullying Detection
- Reinforcement Learning in Robotics
- Advanced Neural Network Applications
- Social Media and Politics
- Topic Modeling
- Rough Sets and Fuzzy Logic
- SARS-CoV-2 detection and testing
- Multimodal Machine Learning Applications
- Web Data Mining and Analysis
- Personal Information Management and User Behavior
- Context-Aware Activity Recognition Systems
- Geoscience and Mining Technology
- Robotics and Automated Systems
- Complex Network Analysis Techniques
- Mental Health via Writing
Georgia Institute of Technology
2023-2025
Microsoft Research (India)
2023
Delhi Technological University
2020
Code-switching in linguistically diverse, low resource languages is often semantically complex and lacks sophisticated methodologies that can be applied to real-world data for precisely detecting hate speech. In an attempt bridge this gap, we introduce a three-tier pipeline employs profanity modeling, deep graph embeddings, author profiling retrieve instances of speech Hindi-English code-switched language (Hinglish) on social media platforms like Twitter. Through extensive comparison against...
For the past few decades, machines have replaced humans in several disciplines. However, machine cognition still lags behind human capabilities. We address machines' ability to recognize drawn sketches this work. Visual representations, such as long been a medium of communication for humans. artificially intelligent systems effectively immerse interactive environments, it is required that understand notations. The abstract nature and varied artistic styling these make automatic recognition...
Suggestion mining tasks are often semantically complex and lack sophisticated methodologies that can be applied to real-world data. The presence of suggestions across a large diversity domains the absence labelled balanced datasets render this task particularly challenging deal with. In an attempt overcome these challenges, we pro-pose two-tier pipeline leverages Discourse Marker based oversampling fine-grained suggestion techniques retrieve from online forums. Through extensive comparison...
We present HyWay, short for "Hybrid Hallway", to enable mingling and informal interactions among physical virtual users, in casual spaces settings, such as office water cooler areas, conference hallways, trade show floors, more. call out how the hybrid unstructured (or semi-structured) nature of settings set these apart from all-virtual and/or structured considered prior work. Key design HyWay is bridging awareness gap between providing users same agency users. To this end, we have designed...
This paper introduces a novel approach to leverage the generalizability capability of Diffusion Models for Source-Free Domain Adaptation (DM-SFDA). Our proposed DM-SFDA method involves fine-tuning pre-trained text-to-image diffusion model generate source domain images using features from target guide process. Specifically, is fine-tuned samples that minimize entropy and maximize confidence model. We then apply established unsupervised adaptation techniques align generated with data. validate...
Over the past few years, Text-to-Image (T2I) generation approaches based on diffusion models have gained significant attention. However, vanilla often suffer from spelling inaccuracies in text displayed within generated images. The capability to generate visual is crucial, offering both academic interest and a wide range of practical applications. To produce accurate images, state-of-the-art techniques adopt glyph-controlled image approach, consisting layout generator followed by an that...
Suggestion mining tasks are often semantically complex and lack sophisticated methodologies that can be applied to real-world data. The presence of suggestions across a large diversity domains the absence labelled balanced datasets render this task particularly challenging deal with. In an attempt overcome these challenges, we propose two-tier pipeline leverages Discourse Marker based oversampling fine-grained suggestion techniques retrieve from online forums. Through extensive comparison on...
Using egocentric video and head motion data from 67 order picking tasks (244 picks;149 orders), we learn visual models of the 10 objects picked to fulfill orders. Boundary segmentations four actions (pick, carry, place, carry empty) had an average test RMSE 1.11 seconds using computer vision 5.53 only $( \approx 39.8$ seconds/task). The were clustered with 93.8% accuracy weak supervision provided by picks (which could occur in any order) specified tasks. We apply resulting on independent...
Domain Adaptation (DA) is a method for enhancing model's performance on target domain with inadequate annotated data by applying the information model has acquired from related source sufficient labeled data. The escalating enforcement of data-privacy regulations like HIPAA, COPPA, FERPA, etc. have sparked heightened interest in adapting models to novel domains while circumventing need direct access data, problem known as Source-Free (SFDA). In this paper, we propose framework SFDA that...
Practical Imitation Learning (IL) systems rely on large human demonstration datasets for successful policy learning. However, challenges lie in maintaining the quality of collected data and addressing suboptimal nature some demonstrations, which can compromise overall dataset hence learning outcome. Furthermore, intrinsic heterogeneity behavior produce equally but disparate further exacerbating challenge discerning quality. To address these challenges, this paper introduces to Discern (L2D),...
Active Learning is a very common yet powerful framework for iteratively and adaptively sampling subsets of the unlabeled sets with human in loop goal achieving labeling efficiency. Most real world datasets have imbalance either classes slices, correspondingly, parts dataset are rare. As result, there has been lot work designing active learning approaches mining these rare data instances. assume access to seed set instances which contain However, event more extreme rareness, it reasonable...