SERB Startup Research Grant
SERB Funded Project [Jan 2024 - Present]
Science and Engineering Research Board
(Statutory Body Established Through an Act of Parliament: SERB Act 2008)
Government of India
Project Details:
- Project Title: Machine Unlearning for Selective Removal of Digital Data Footprint from Deep Learning Models
- File Number: SRG/2023/001686
- Principal Investigator (PI): Dr. Murari Mandal
- Institute: Kalinga Institute of Industrial Technology (KIIT)
- Project Start Date: 02-Jan-2024
- Duration: 24 months
- Project Completion Date: 01-Jan-2026
- Total Budget: 30,50,300 INR
- Status: Ongoing
- Budget Released So Far: 23,30,000 INR
- Research Fellow: Umakanta Maharana
Project Overview
Our project, supported by the Science and Engineering Research Board (SERB), aims to address one of the most pressing concerns in today’s digital age: the selective removal of digital data footprints from deep learning models. In the rapidly evolving landscape of artificial intelligence, the ability to unlearn specific data points from a trained model is crucial. This capability is essential for complying with privacy regulations, such as the GDPR, and for ensuring that models do not perpetuate biases or inaccuracies associated with certain data. Some related papers:
- Can Bad Teaching Induce Forgetting? Unlearning in Deep Networks Using an Incompetent Teacher
- Fast Yet Effective Machine Unlearning
Objectives
The primary goal of this project is to develop methodologies for effectively and efficiently removing specific data from deep learning models without compromising their overall performance. This involves:
- Designing algorithms for selective data removal.
- Ensuring the model’s robustness and accuracy post-unlearning.
- Addressing the challenges of data integrity and consistency.
Current Progress
- Algorithm Development: Initial prototypes of unlearning algorithms have been tested with promising results.
- Model Training: Deep learning models have been trained with diverse datasets to evaluate the effectiveness of the unlearning process.
- Budget Utilization: Out of the total budget of 30,50,300 INR, 23,30,000 INR has been released and utilized effectively to support research activities.
Publications Supported by the Project
- Ayush K Tarun, Vikram S Chundawat, Murari Mandal, and Hong Ming Tan, Bowei Chen, Mohan Kankanhalli, “EcoVal: An Efficient Data Valuation Framework for Machine Learning”, KDD, Barcelona, 2024 [Core A*]
- Aakash Sen Sharma, Niladri Sarkar, Vikram Chundawat, Ankur A Mali, Murari Mandal, “Unlearning or Concealment? A Critical Analysis and Evaluation Metrics for Unlearning in Diffusion Models”, Under Review (AAAI-24), Paper available on Arxiv
- Romit Chatterjee, Vikram Chundawat, Ayush Tarun, Ankur Mali, Murari Mandal, “A Unified Framework for Continual Learning and Machine Unlearning”, Under Review (AAAI-24), Paper available on Arxiv
For more information and updates on this project, feel free to reach out to me at murari.nus@gmail.com.