A5065006343- Topic Modeling
- Natural Language Processing Techniques
- Human Pose and Action Recognition
- Stock Market Forecasting Methods
- Advanced Text Analysis Techniques
- Multimodal Machine Learning Applications
- Semantic Web and Ontologies
- Complex Systems and Time Series Analysis
- Handwritten Text Recognition Techniques
- Credit Risk and Financial Regulations
- Anomaly Detection Techniques and Applications
- Complex Network Analysis Techniques
- Gait Recognition and Analysis
- Image Retrieval and Classification Techniques
- Financial Markets and Investment Strategies
- Bayesian Methods and Mixture Models
- Sentiment Analysis and Opinion Mining
- Video Surveillance and Tracking Methods
- Data Mining Algorithms and Applications
- Sustainable Finance and Green Bonds
- Generative Adversarial Networks and Image Synthesis
- Statistical Methods and Inference
- Music and Audio Processing
- Explainable Artificial Intelligence (XAI)
- Video Analysis and Summarization
Harvard University
2025
Morgan Stanley (United Kingdom)
2022-2023
JPMorgan Chase & Co (United States)
2022
Georgia Institute of Technology
2022
IBM Research - Almaden
2021
California University of Pennsylvania
2021
University of Colorado System
2021
Hong Kong University of Science and Technology
2020
University of Hong Kong
2020
Carleton College
2020
The Fugl Meyer Assessment (FMA) is a widely-used assessment for tracking motor function recovery post-stroke. Due to the limited access rehabilitation, there exists need remote and automated solutions. Wearable sensors data-driven methods have shown promise enabling automatic upper extremity FMA (FMA-UE) estimation, but minimizing user input motion aligning with current clinical activities will aid adoption of sensor-based assessments. In this work, we present an FMA-UE estimator which can...
Raj Shah, Kunal Chawla, Dheeraj Eidnani, Agam Wendi Du, Sudheer Chava, Natraj Raman, Charese Smiley, Jiaao Chen, Diyi Yang. Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. 2022.
Synthetic data has made tremendous strides in various commercial settings including finance, healthcare, and virtual reality. We present a broad overview of prototypical applications synthetic the financial sector particular provide richer details for few select ones. These cover wide variety modalities tabular, time-series, event-series, unstructured arising from both markets retail applications. Since finance is highly regulated industry, potential approach dealing with issues related to...
The integration of Environmental, Social and Governance (ESG) considerations into business decisions investment strategies have accelerated over the past few years. It is important to quantify extent which ESG-related conversations are carried out by companies so that their impact on operations can be objectively assessed. However, profiling ESG language challenging due its multi-faceted nature lack supervised datasets. This research study aims detect historical trends in discussions...
Enterprise documents such as forms, invoices, receipts, reports, contracts, and other similar records, often carry rich semantics at the intersection of textual spatial modalities. The visual cues offered by their complex layouts play a crucial role in comprehending these effectively. In this paper, we present DocLLM, lightweight extension to traditional large language models (LLMs) for reasoning over documents, taking into account both layout. Our model differs from existing multimodal LLMs...
We present a contrastive study of document-level event classification range seven different types, namely floods, storms, fires, armed conflict, terrorism, infrastructure breakdown and labour unavailability from English-language news. Our compares supervised approaches, Support Vector Machine (SVM), Random Forest (RF), Convolutional Neural Network (CNN) Hierarchical Attention (HAN). While past systems for Topic Detection Tracking (TDT) extraction have proposed machine learning models, to...
The automatic extraction of breaking news events from natural language text is a valuable capability for decision support systems. Traditional systems tend to focus on extracting single media source and often ignore cross-media references. Here, we describe large-scale automated system disasters critical both newswire social media. We outline comprehensive architecture that can identify, categorize summarize seven different event types - namely floods, storms, fires, armed conflict,...
As large language models (LLMs) expand the power of natural processing to handle long inputs, rigorous and systematic analyses are necessary understand their abilities behavior. A salient application is summarization, due its ubiquity controversy (e.g., researchers have declared death summarization). In this paper, we use financial report summarization as a case study because reports not only but also numbers tables extensively. We propose computational framework for characterizing...
Graph-structured diagrams, such as enterprise ownership charts or management hierarchies, are a challenging medium for deep learning models they not only require the capacity to model language and spatial relations but also topology of links between entities varying semantics what those represent. Devising Question Answering that automatically process understand diagrams have vast applications many domains, can move state-of-the-art on multimodal document understanding new frontier. Curating...
Conditional Random Fields (CRF), a structured prediction method, combines probabilistic graphical models and discriminative classification techniques in order to predict class labels sequence recognition problems. Its extension the Hidden (HCRF) uses hidden state variables capture intermediate structures. The number of states an HCRF must be specified priori. This is often not known advance. A non-parametric HCRF, with automatically inferred from data, proposed here. significant advantage...
Retrieving relevant documents from a corpus is typically based on the semantic similarity between document content and query text. The inclusion of structural relationship can benefit retrieval mechanism by addressing gaps. However, incorporating these relationships requires tractable mechanisms that balance structure with semantics take advantage prevalent pre-train/fine-tune paradigm. We propose here holistic approach to learning representations integrating intra-document inter-document...
Price evaluations of municipal bonds have traditionally been performed by human experts based on their market knowledge and trading experience. Automated evaluation is an attractive alternative providing the advantage objective estimation that transparent, consistent, scalable. In this paper, we present a statistical model to automatically estimate U.S bond yields trade transactions study agreement between machine generated estimates. The uses piecewise polynomials constructed using basis...
When analyzing companies in financial markets, it is essential to identify those that share similar characteristics order assess their relative strengths and weaknesses. This challenging task requires representing the rich set of information associated with complex interrelations between them a form amenable pattern recognition. We present here new deep representation learning method encodes network graph low-dimensional embedding space, preserving its topological structure. Our solution...
Information visualization is critical to analytical reasoning and knowledge discovery. We present an interactive studio that integrates perceptive techniques with powerful text analytics algorithms assist humans in sense-making of large complex corpora. The novel visual representations introduced here encode the features delivered by modern mining models using advanced metaphors such as hypergraphs, nested topologies tessellated planes. They enhance human-computer interaction experience for...
Large language models (LLMs) are primarily designed to understand unstructured text. When directly applied structured formats such as tabular data, they may struggle discern inherent relationships and overlook critical patterns. While representation learning methods can address some of these limitations, existing efforts still face challenges with sparse high-cardinality fields, precise numerical reasoning, column-heavy tables. Furthermore, leveraging learned representations for downstream...
Graph Neural Networks (GNNs) have been widely deployed in various real-world applications. However, most GNNs are black-box models that lack explanations. One strategy to explain is through counterfactual explanation, which aims find minimum perturbations on input graphs change the GNN predictions. Existing works explanations primarily concentrate local-level perspective (i.e., generating counterfactuals for each individual graph), suffers from information overload and lacks insights into...
Large Language Models (LLMs) are highly resource-intensive to fine-tune due their enormous size. While low-rank adaptation is a prominent parameter-efficient fine-tuning approach, it suffers from sensitivity hyperparameter choices, leading instability in model performance on downstream tasks. This paper highlights the importance of effective parameterization reduce estimator variance and enhance stability final outputs. We propose MonteCLoRA, an efficient technique, employing Monte Carlo...