Arka Dutta

Work Experience

• Rochester Institute of Technology, Rochester, NY
  Graduate Research Assistant (Aug. 2023 – Present)
  • PhD student advised by Prof. Dr. Ashiqur R. KhudaBukhsh. Research focus on Responsible AI and equitable AI systems.
  • Published papers in A* conference such as IJCAI; research covered in widely circulated AI Magazines like Montreal AI Ethics.
• Rochester Institute of Technology, Remote
  Research Intern (Sep. 2022 – Feb. 2023)
  • Supervised by Prof. Dr. Ashiqur R. KhudaBukhsh: Proposed a novel research direction on web platform preservation and web archaeology through platform-level and user-level analysis.
  • Conducted a thorough analysis of temporal and socio-political trends in approximately 150 million user comments, derived from real-life news sources (YouTube news channels), spanning over a decade of natural language data.
• Indian Statistical Institute, Kolkata, India
  NLP Research Intern (Jul. 2022 – Oct. 2022)
  • Supervised by Prof. Dr. Utpal Garain: Devised an approach to extract hate-phrases for hate-speech classification in Bengali and Hindi, two low-resource languages using HASOC dataset.
  • Used Language Models like RoBERTa, AlBERT and XAI methods like Deletion-intervention, game theory-based Feature Ablation to extract hate speech that resulted in identifying ~73% hate-phrases towards explaining the final classifier prediction. This resulted in a ~12% improvement in classification accuracy in hate-speech classification than the previous baseline.
• Jadavpur University, Remote
  Research Intern (Dec. 2021 – Jul. 2022)
  • Under the supervision of Dr. Arnab Raha, Staff Research Scientist, Intel AI: Systemized a Distributed Deep Neural Network approach with Adaptive Workload Partitioning along low-computing edge devices to minimize the latency due to cloud transfer for faster and resource-friendly computation and communication.
  • Used Neural Network based optimization to achieve a 2.3x dip in latency improving computation efficiency, which could translate into significant cost savings for edge computing applications.