Uncovering Conceptual Blindspots in Generative Image Models Using Sparse Autoencoders
This addresses the issue of understanding structural limitations in generative AI for researchers and practitioners, though it is incremental as it builds on existing sparse autoencoder techniques.
The paper tackles the problem of generative image models failing to produce images with simple concepts present in training data, such as human hands or groups of four objects, by introducing a systematic method using sparse autoencoders to identify and characterize these conceptual blindspots. The result includes the discovery of specific suppressed blindspots like bird feeders and DVD discs, and exaggerated ones like wood textures, across models like Stable Diffusion and PixArt.
Despite their impressive performance, generative image models trained on large-scale datasets frequently fail to produce images with seemingly simple concepts -- e.g., human hands or objects appearing in groups of four -- that are reasonably expected to appear in the training data. These failure modes have largely been documented anecdotally, leaving open the question of whether they reflect idiosyncratic anomalies or more structural limitations of these models. To address this, we introduce a systematic approach for identifying and characterizing "conceptual blindspots" -- concepts present in the training data but absent or misrepresented in a model's generations. Our method leverages sparse autoencoders (SAEs) to extract interpretable concept embeddings, enabling a quantitative comparison of concept prevalence between real and generated images. We train an archetypal SAE (RA-SAE) on DINOv2 features with 32,000 concepts -- the largest such SAE to date -- enabling fine-grained analysis of conceptual disparities. Applied to four popular generative models (Stable Diffusion 1.5/2.1, PixArt, and Kandinsky), our approach reveals specific suppressed blindspots (e.g., bird feeders, DVD discs, and whitespaces on documents) and exaggerated blindspots (e.g., wood background texture and palm trees). At the individual datapoint level, we further isolate memorization artifacts -- instances where models reproduce highly specific visual templates seen during training. Overall, we propose a theoretically grounded framework for systematically identifying conceptual blindspots in generative models by assessing their conceptual fidelity with respect to the underlying data-generating process.