I am a final year undergraduate student at the department of Computer Science and Engineering at Indian Institute of Technology, Kharagpur (IIT KGP). I am currently interning remotely under the guidance of Dr. Nafise Moosavi and Dr. Gozde Gul Sahin of the Ubiquitous Knowledge Processing (UKP) lab at TU Darmstadt. In the past I have had the opportunity to work under the guidance of Prof. Lili Mou from the University of Alberta, Dr. Niyati Chhaya and Suryateja Bulusu from Adobe Research, India and Rajdeep Mukherjee a research scholar and PhD candidate at the Complex Networks Research Group at IIT KGP.
My primary interests lie in the field of Natural Language Processing (NLP). In the field of NLP I am primarly interested in explaiable deep learning, bias and artefact detection, natural language inference (NLI), natural language generation and dialogue.
Till now I've had the opportunity to work on summarization, hate speech detection, headline generation, emotion detection, multi-lingual dialogue and bias/artefact detection. In my Adobe internship I had the unique opportunity of working on a project that lay in the intersection of human computer interaction, reinforcement learning and computer vision.
I wish to pursue NLP research as a masters (and potentially a PhD) student. My dream goal would be to successfully apply NLP towards creating intelligent assistants for creative writing, like screenwriting, storyboarding, comic book creation etc. and towards using social media to gather clues to locate missing people. I also wish to tackle these issues with explainable models that can justify their output. For e.g. a scriptwriting assistant, while suggesting an edit, should give a rationale like, "this makes the main character more compelling and relatable". In case of a system to locate missng people, if the system suggests that the public interaction between two people is suspicious, then it should give probable cause which could be used by the investigating agency to subpoena the social media platform for additional details (like chatlogs or other private information). This would help investigators to filter out spurious leads and save time.
List of publications I have co-authored, including those that have been submitted to a conference and are currently under review.
Understanding the Role of Affect Dimensions in Detecting Emotions from Tweets: A Multi-task Approach (SIGIR 2021)
Rajdeep Mukherjee, Atharva Naik, Sriyash Poddar, Soham Dasgupta, Niloy GangulyWe propose VADEC, a multi-task framework that exploits the correlation between the categorical and dimensional models of emotion representation for better subjectivity analysis. Focusing primarily on the effective detection of emotions from tweets, we jointly train multi-label emotion classification and multi-dimensional emotion regression, thereby utilizing the inter-relatedness between the tasks. Co-training especially helps in improving the performance of the classification task as we outperform the strongest baselines with 3.4%, 11%, and 3.9% gains in Jaccard Accuracy, Macro-F1, and Micro-F1 scores respectively on the AIT dataset. We also achieve state-of-the-art results with 11.3% gains averaged over six different metrics on the SenWa...
Weakly Supervised Explainable Phrasal Reasoning with Neural Fuzzy Logic (Under Review)
Zijun Wu, Atharva Naik, Zi Xuan Zhang, Lili MouNatural language inference (NLI) aims to determine the logical relationship between two sentences among the target labels Entailment, Contradiction, and Neutral. In recent years, deep learning models have become a prevailing approach to NLI, but they lack interpretability and explainability. In this work, we address the explainability for NLI by weakly supervised logical reasoning, and propose an Explainable Phrasal Reasoning (EPR) approach. Our model first detects phrases as the semantic unit and aligns corresponding phrases. Then, the model predicts the NLI label for the aligned phrases, and induces the sentence label by fuzzy logic formulas. Our EPR is almost everywhere differentiable and thus the system can be trained end-to-end in a weakly supervised manner. We annotated a corpus and developed a set of metrics to evaluate phrasal reasoning. Results show that our EPR yields much more meaningful explanations in terms of F scores than previous studies. To the best of our knowledge, we are the first to develop a weakly...
Representation Learning for Conversational Data using Discourse Mutual Information Maximization (Under Review)
Bishal Santra, Sumegh Roychowdhury, Aishik Mandal, Vasu Gurram, Atharva Naik, Manish Gupta, Pawan GoyalCurrent methods for pretraining natural language representations typically ignore any information available through the discourse level structures in a text. Such structures may have important information, useful specifically for dialog understanding, discourse segmentation and many other domain specific tasks. To capture such features at the pretraining stage itself, we propose an information theoretic objective for learning discourse level features. Through an extensive experimental study, we show that representations learned using the proposed objectives has better understanding of full dialog contexts than existing methods. Additionally, our models show strong performance in utterance level dialog understanding tasks like Intent Detection or Act classification. These proposed mutual information based objectives are very effective learners. Small scale...
CollabColor: Creative Support System for Human-Human Synchronous Collaboration (Under Review)
Suryateja BV, Atharva Naik, Yash Parag Butala, Jeet Patel, Sristy Sharma, Niyati ChhayaHumans are unique in working collaboratively by sharing and understanding intentions. However, digital collaboration is daunting, especially in creative design life cycles, due to non-linear workflows and lack of micro-alignments coupled with the need for robust network connectivity. We present a formative study with creatives to identify key themes in conflicts that arise in this space. We introduce CollabColor, an intelligent system that aids in resolving conflicts for two users synchronously collaborating on a low-touch creative task. More specifically, given an uncolored line-art on a canvas and a set of reference images from the users as input, we provide non-obtrusive interventions during their real-time collaboration to ensure that the final colorization of the art is coherent, and all the users’ aligned preferences are...
My Github Statistics
I am an open source enthusiast. I love Python, and have worked on a wide variety of projects using Python. I have experience with Deep learning frameworks like PyTorch, tensorflow, keras, sklearn etc., NLP toolkits like NLTK, spacy and gensim and data science and scientific computing tools like pandas, jupyter, numpy and scipy.
I also have extensive experience with web scraping and automation using beautiful soup and selenium
I love to tinker with tkinter (pun intended )) and GUI programming in PyQt5. I like building UI tools that are useful to me using PyQt5, in my free time.
I am also proficient in C, C++ and bash scripting. I have also worked with the CUDA API/extension of C++ in the past. I have some basic understanding of Apache Spark programming with scala.
Hobbies and Interests
I absolutely love Anime, True Crime, Script Writing and learning new languages
Get in touch with me
Click the button below to open the contact form. This 'contact' form is just a colorlib template that I stole and the emailing mechanism is email.js. Don't judge me ok!