Author Archives: jkh6

Self-Improving Diffusion Models with Synthetic Data

Sina Alemohammad, Ahmed Imtiaz Humayun, Richard Baraniuk
Rice University
Shruti Agarwal, John Collomosse
Adobe Research

arxiv.org/abs/2408.16333, 30 August 2024

Abstract: The artificial intelligence (AI) world is running out of real data for training increasingly large generative models, resulting in accelerating pressure to train on synthetic data. Unfortunately, training new generative models with synthetic data from current or past generation models creates an autophagous (self-consuming) loop that degrades the quality and/or diversity of the synthetic data in what has been termed model autophagy disorder (MAD) and model collapse. Current thinking around model autophagy recommends that synthetic data is to be avoided for model training lest the system deteriorate into MADness. In this paper, we take a different tack that treats synthetic data differently from real data. Self-IMproving diffusion models with Synthetic data (SIMS) is a new training concept for diffusion models that uses self-synthesized data to provide negative guidance during the generation process to steer a model's generative process away from the non-ideal synthetic data manifold and towards the real data distribution. We demonstrate that SIMS is capable of self-improvement; it establishes new records based on the Fréchet inception distance (FID) metric for CIFAR-10 and ImageNet-64 generation and achieves competitive results on FFHQ-64 and ImageNet-512. Moreover, SIMS is, to the best of our knowledge, the first prophylactic generative AI algorithm that can be iteratively trained on self-generated synthetic data without going MAD. As a bonus, SIMS can adjust a diffusion model's synthetic data distribution to match any desired in-domain target distribution to help mitigate biases and ensure fairness.

The figure above illustrates that SIMS simultaneously improves diffusion modeling and synthesis performance while acting as a prophylactic against Model Autophagy Disorder (MAD). First row: Samples from a base diffusion model (EDM2-S) trained on 1.28M real images from the ImageNet-512 dataset (Fréchet inception distance, FID = 2.56). Second row: Samples from the base model after fine-tuning with 1.5M images synthesized from the base model, which degrades synthesis performance and pushes the model towards MADness (model collapse) (FID = 6.07). Third row: Samples from the base model after applying SIMS using the same self-generated synthetic data as in the second row (FID = 1.73).

The U.S. National Science Foundation announced today a strategic investment of $90 million over five years in SafeInsights, a unique national scientific cyberinfrastructure aimed at transforming learning research and STEM education. Funded through the Mid-Scale Research Infrastructure Level-2 program (Mid-scale RI-2), SafeInsights is led by Prof. Richard Baraniuk at OpenStax at Rice University, who will oversee the implementation and launch of this new research infrastructure project of unprecedented scale and scope.

SafeInsights aims to serve as a central hub, facilitating research coordination and leveraging data across a range of major digital learning platforms that currently serve tens of millions of U.S. learners across education levels and science, technology, engineering and mathematics.

With its controlled and intuitive framework, unique privacy-protecting approach and emphasis on the inclusion of students, educators and researchers from diverse backgrounds, SafeInsights will enable extensive, long-term research on the predictors of effective learning, which are key to academic success and persistence.

Links for more information:

Two DSP group papers have been accepted by the International Conference on Learning Representations (ICLR) 2024 in Vienna, Austria

Self-Consuming Generative Models Go MAD
http://arxiv.org/abs/2307.01850

Sina Alemohammad, Josue Casco-Rodriguez, Lorenzo Luzi, Ahmed Imtiaz Humayun,
Hossein Babaei, Daniel LeJeune, Ali Siahkoohi, Richard G. Baraniuk

Abstract:  Seismic advances in generative AI algorithms for imagery, text, and other data types has led to the temptation to use synthetic data to train next-generation models. Repeating this process creates an autophagous ("self-consuming") loop whose properties are poorly understood. We conduct a thorough analytical and empirical analysis using state-of-the-art generative image models of three families of autophagous loops that differ in how fixed or fresh real training data is available through the generations of training and in whether the samples from previous-generation models have been biased to trade off data quality versus diversity. Our primary conclusion across all scenarios is that without enough fresh real data in each generation of an autophagous loop, future generative models are doomed to have their quality (precision) or diversity (recall) progressively decrease. We term this condition Model Autophagy Disorder (MAD), making analogy to mad cow disease.

In the news:

In cartoons:

Rice DSP graduate student Jack Wang successfully defended his PhD thesis entitled "Towards Personalized Human Learning at Scale: A Machine Learning Approach."

Abstract: Despite the recent advances in artificial intelligence (AI) and machine learning (ML), we have yet to witness the transformative breakthroughs they can bring to education and, more broadly, to how humans learn. This thesis establishes two research directions that leverage the recent advances in generative modeling to enable more personalized learning experiences on a large scale. The first part of the thesis focuses on educational content generation and proposes a method to automatically generate math word problems that are personalized to each learner. The second part of the thesis focuses on learning analytics and proposes a framework for analyzing learners’ open-ended solutions to assessment questions, such as code submissions in computer science education.

Jack’s next step is Adobe Research, where he will be working on new natural language processing models for documents and other data.