Mitigating Biases in Self-consuming Generative Models

#Artificial_Intelligence, #Reliability, #Generative_Models, #Self_Consumption, #Computer_Science,
Share

IEEE CS SD Invited Seminar Series 2025 - Lecture 8 (Virtual)

 


In this talk, Dr. Ali Siahkoohi highlights the risks of the current industrial AI practices involving training large-scale generative models on vast amounts of data scraped from the internet. This process unwittingly leads to training newer models on increasing amounts of AI-synthesized data that is rapidly proliferating online, a phenomenon Dr. Siahkoohi refers to as ``model autophagy'' (self-consuming models). He shows that without a sufficient influx of fresh, real data at each stage of an autophagous loop, future generative models will inevitably suffer a decline in either quality (precision) or diversity (recall). To mitigate this issue and inspired by fixed-point optimization, a penalty to the loss function of generative models is introduced that minimizes discrepancies between the model's weights when trained on real versus synthetic data. Since computing this penalty would require training a new generative model at each iteration, a permutation-invariant hypernetwork is proposed to make evaluating the penalty tractable by dynamically mapping data batches to model weights. This ensures scalability and seamless integration of the penalty term into existing generative modeling paradigms, mitigating biases associated with model autophagy. Additionally, this penalty improves the representation of minority classes in imbalanced datasets, which is a key step toward enhancing fairness in generative models.



  Date and Time

  Location

  Hosts

  Registration



  • Date: 11 Jun 2025
  • Time: 12:30 AM UTC to 01:30 AM UTC
  • Add_To_Calendar_icon Add Event to Calendar
If you are not a robot, please complete the ReCAPTCHA to display virtual attendance info.
  • Contact Event Host
  • Co-sponsored by Media Partner: Open Research Institute (ORI)
  • Starts 27 May 2025 07:00 AM UTC
  • Ends 11 June 2025 02:00 AM UTC
  • No Admission Charge


  Speakers

Ali Siahkoohi of University of Central Florida

Topic:

Mitigating Biases in Self-consuming Generative Models

Biography:

Ali Siahkoohi is an incoming tenure-track assistant professor in University of Central Florida's Computer Science Department. Currently, he is a Simons Postdoctoral Fellow in the Department of Computational Applied Mathematics & Operations Research at Rice University, jointly hosted by Dr. Maarten V. de Hoop and Dr. Richard G. Baraniuk. He received his Ph.D. in Computational Science and Engineering from Georgia Institute of Technology in 2022. His research focuses on designing scalable methods for quantifying uncertainty in AI models, with a broader goal of enhancing AI reliability.





Agenda

  • Invited talk from Dr. Ali Siahkoohi, Assistant Professor in University of Central Florida's Computer Science Department.
  • Q/A Session


- 8th lecture of the 2025 Invited Seminar Series (Virtual) organized by IEEE Computer Society San Diego Chapter. Previous lectures: 2023, 2024, and 2025 invited seminar series.