A5060923104- Topic Modeling
- Natural Language Processing Techniques
- Misinformation and Its Impacts
- Speech and dialogue systems
- Biomedical Text Mining and Ontologies
- Sentiment Analysis and Opinion Mining
- Semantic Web and Ontologies
- Digital Communication and Language
- Language, Discourse, Communication Strategies
- Language, Metaphor, and Cognition
- Photoacoustic and Ultrasonic Imaging
- Image and Signal Denoising Methods
- Advanced Text Analysis Techniques
- Masonry and Concrete Structural Analysis
- Authorship Attribution and Profiling
- Multimodal Machine Learning Applications
- Structural Response to Dynamic Loads
- Optical Coherence Tomography Applications
- Fuzzy Logic and Control Systems
- Advanced Image Fusion Techniques
- Land Use and Ecosystem Services
- Digital Games and Media
- Advanced Graph Neural Networks
- 3D Surveying and Cultural Heritage
- Multi-Agent Systems and Negotiation
Shri Ramswaroop Memorial University
2019-2024
Carnegie Mellon University
2018-2023
Pacific Northwest National Laboratory
2023
Northwestern University
2019
Birla Institute of Technology and Science, Pilani
2018
Indian Institute of Science Bangalore
2018
Distantly-supervised Relation Extraction (RE) methods train an extractor by automatically aligning relation instances in a Knowledge Base (KB) with unstructured text. In addition to instances, KBs often contain other relevant side information, such as aliases of relations (e.g., founded and co-founded are for the founderOfCompany). RE models usually ignore readily available information. this paper, we propose RESIDE, distantly-supervised neural extraction method which utilizes additional...
Conditional language models are predominantly trained with maximum likelihood estimation (MLE), giving probability mass to sparsely observed target sequences. While MLE assign high plausible sequences given the context, model probabilities often do not accurately rank-order generated by quality. This has been empirically in beam search decoding as output quality degrading large sizes, and strategies benefiting from heuristics such length normalization repetition-blocking. In this work, we...
Biomedical natural language processing tools are increasingly being applied for broad-coverage information extraction-extracting medical of all types in a scientific document or clinical note. In such settings, linking mentions concepts to standardized vocabularies requires choosing the best candidate from large inventories covering dozens types. This study presents novel semantic type prediction module biomedical NLP pipelines and two automatically-constructed, large-scale datasets with...
Twitter has become one of the most sought after places to discuss a wide variety topics, including medically relevant issues such as cancer. This helps spread awareness regarding various causes, cures and prevention methods However, no proper analysis been performed, which discusses validity claims. In this work, we aim tackle misinformation in platforms. We collect present dataset tweets talk specifically about cancer propose an attention-based deep learning model for automated detection...
To successfully negotiate a deal, it is not enough to communicate fluently: pragmatic planning of persuasive negotiation strategies essential. While modern dialogue agents excel at generating fluent sentences, they still lack grounding and cannot reason strategically. We present DialoGraph, system that incorporates in using graph neural networks. DialoGraph explicitly dependencies between sequences enable improved interpretable prediction next optimal strategies, given the context. Our...
Learning from human feedback has been shown to be effective at aligning language models with preferences. Past work often relied on Reinforcement Human Feedback (RLHF), which optimizes the model using reward scores assigned a trained preference data. In this we show how recently introduced Sequence Likelihood Calibration (SLiC), can also used effectively learn preferences (SLiC-HF). Furthermore, demonstrate done data collected for different model, similar off-policy, offline RL Automatic and...
Improving the alignment of language models with human preferences remains an active research challenge. Previous approaches have primarily utilized Reinforcement Learning from Human Feedback (RLHF) via online RL methods such as Proximal Policy Optimization (PPO). Recently, offline Sequence Likelihood Calibration (SLiC) and Direct Preference (DPO) emerged attractive alternatives, offering improvements in stability scalability while maintaining competitive performance. SLiC refines its loss...
Recent studies have shown that large language models' (LLMs) mathematical problem-solving capabilities can be enhanced by integrating external tools, such as code interpreters, and employing multi-turn Chain-of-Thought (CoT) reasoning. While current methods focus on synthetic data generation Supervised Fine-Tuning (SFT), this paper the complementary direct preference learning approach to further improve model performance. However, existing algorithms are originally designed for single-turn...
Ritam Dutt, Sayan Sinha, Rishabh Joshi, Surya Shekhar Chakraborty, Meredith Riggs, Xinru Yan, Haogang Bao, Carolyn Rose. Proceedings of the 16th Conference European Chapter Association for Computational Linguistics: Main Volume. 2021.
Keyphrase extraction aims at automatically extracting a list of "important" phrases representing the key concepts in document. Prior approaches for unsupervised keyphrase resorted to heuristic notions phrase importance via embedding clustering or graph centrality, requiring extensive domain expertise. Our work presents simple alternative approach which defines keyphrases as document that are salient predicting topic To this end, we propose INSPECT—an uses self-explaining models identifying...
Modern Massively Multi-player Online Games (MMOGs) have grown to become extremely complex in terms of the usable resources games, resulting an increase amount data collected by tracking in-game activities players. This has opened door for researchers come up with novel methods utilize this improve and personalize user experience. In paper, a but flexible framework towards building team based recommender system player-versus-player (PvP) content such MMOGs is presented, applied case study...
In this paper we describe our submission for the task of Propaganda Span Identification in news articles. We introduce a BERT-BiLSTM based span-level propaganda classification model that identifies which token spans within sentence are indicative propaganda. The ”multi-granular” incorporates linguistic knowledge at various levels text granularity, including word, and document level syntactic, semantic pragmatic affect features, significantly improve performance, compared to its...
The problem of building a coherent and non-monotonous conversational agent with proper discourse coverage is still an area open research. Current architectures only take care semantic contextual information for given query fail to completely account syntactic external knowledge which are crucial generating responses in chit-chat system. To overcome this problem, we propose end multi-stream deep learning architecture learns unified embeddings query-response pairs by leveraging from memory...
The notion of face refers to the public self-image an individual that emerges both from individual’s own actions as well interaction with others. Modeling and understanding its state changes throughout a conversation is critical study maintenance basic human needs in through interaction. Grounded politeness theory Brown Levinson (1978), we propose generalized framework for modeling acts persuasion conversations, resulting reliable coding manual, annotated corpus, computational models....
Aligning language models (LMs) with curated human feedback is critical to control their behaviors in real-world applications. Several recent policy optimization methods, such as DPO and SLiC, serve promising alternatives the traditional Reinforcement Learning from Human Feedback (RLHF) approach. In practice, often comes a format of ranked list over multiple responses amortize cost reading prompt. Multiple can also be by reward or AI feedback. There lacks study on directly fitting upon...
Ensuring alignment of language models' outputs with human preferences is critical to guarantee a useful, safe, and pleasant user experience. Thus, has been extensively studied recently several methods such as Reinforcement Learning from Human Feedback (RLHF), Direct Policy Optimisation (DPO) Sequence Likelihood Calibration (SLiC) have emerged. In this paper, our contribution two-fold. First, we show the equivalence between two recent methods, namely Identity (IPO) Nash Mirror Descent...
The dominant framework for alignment of large language models (LLM), whether through reinforcement learning from human feedback or direct preference optimisation, is to learn data. This involves building datasets where each element a quadruplet composed prompt, two independent responses (completions the prompt) and between responses, yielding preferred dis-preferred response. Such data typically scarce expensive collect. On other hand, \emph{single-trajectory} triplet response naturally more...
Existing preference optimization methods are mainly designed for directly learning from human feedback with the assumption that paired examples (preferred vs. dis-preferred) available. In contrast, we propose a method can leverage unpaired preferred or dis-preferred examples, and works even when only one type of (positive negative) is This flexibility allows us to apply it in scenarios varying forms models, including training generative language models based on as well policies sequential...
Reward models (RMs) play a pivotal role in aligning large language (LLMs) with human preferences. However, traditional RM training, which relies on response pairs tied to specific prompts, struggles disentangle prompt-driven preferences from prompt-independent artifacts, such as length and format. In this work, we expose fundamental limitation of current training methods, where RMs fail effectively distinguish between contextual signals irrelevant artifacts when determining To address this,...