Deep neural networks often forget previously learned information when trained sequentially on new objectives, a phenomenon known as catastrophic forgetting. Existing generative strategies combat this issue by randomly sampling rehearsal examples from a generative model. Such an approach contradicts buffer-based approaches where sampling strategy plays an important role. We propose to bridge this gap and benefit from the combination of DDPM trained on the previous task and the classifier guidance technique to actively generate rehearsal examples specifically designed to minimize forgetting in the currently trained classifier. Our experimental results show that GUIDE significantly reduces catastrophic forgetting, outperforming conventional random sampling approaches and surpassing recent state-of-the-art methods in continual learning with generative replay, and buffer-based rehearsal.

Novel diffusion models can synthesize photo-realistic images with integrated high-quality text. Surprisingly, we demonstrate through attention activation patching that only less than $1$\% of diffusion models' parameters, all contained in attention layers, influence the generation of textual content within the images. Building on this observation, we improve textual generation efficiency and performance by targeting cross and joint attention layers of diffusion models. We introduce several applications that benefit from localizing the layers responsible for textual content generation. We first show that a LoRA-based fine-tuning solely of the localized layers enhances, even more, the general text-generation capabilities of large diffusion models while preserving the quality and diversity of the diffusion models' generations. Then, we demonstrate how we can use the localized layers to edit textual content in generated images. Finally, we extend this idea to the practical use case of preventing the generation of toxic text in a cost-free manner. In contrast to prior work, our localization approach is broadly applicable across various diffusion model architectures, including U-Net (e.g., SDXL and DeepFloyd IF) and transformer-based (e.g., Stable Diffusion 3), utilizing diverse text encoders (e.g., from CLIP to the large language models like T5). Project page available at https://t2i-text-loc.github.io/.

Diffusion models, while powerful, can inadvertently generate harmful or undesirable content, raising significant ethical and safety concerns. Recent machine unlearning approaches offer potential solutions but often lack transparency, making it difficult to understand the changes they introduce to the base model. In this work, we introduce SAeUron, a novel method leveraging features learned by sparse autoencoders (SAEs) to remove unwanted concepts in text-to-image diffusion models. First, we demonstrate that SAEs, trained in an unsupervised manner on activations from multiple denoising timesteps of the diffusion model, capture sparse and interpretable features corresponding to specific concepts. Building on this, we propose a feature selection method that enables precise interventions on model activations to block targeted content while preserving overall performance. Our evaluation shows that SAeUron outperforms existing approaches on the UnlearnCanvas benchmark for concepts and style unlearning, and effectively eliminates nudity when evaluated with I2P. Moreover, we show that with a single SAE, we can remove multiple concepts simultaneously and that in contrast to other methods, SAeUron mitigates the possibility of generating unwanted content under adversarial attack. Code and checkpoints are available at GitHub.

Motivation: Neural networks suffer from abrupt loss in performance when retrained with additional data from different distributions. At the same time, training with additional data without access to the previous examples rarely improves the model’s performance. Methods: We propose Adapt & Align, a novel continual learning framework that leverages generative models to align their latent representations across tasks. The approach is divided into two phases: • Local Training: Train a generative model (e.g., a Variational Autoencoder (VAE) or a Generative Adversarial Network (GAN)) on the current task to capture task-specific features. • Global Training: Use a translator network to map these task-specific latent representations into a unified global latent space, thereby facilitating both forward and backward knowledge transfer. Results: Experiments on benchmark datasets (e.g., MNIST, Omniglot, CIFAR, CelebA) as well as real-world application for particle simulation at CERN demonstrate that Adapt & Align mitigates catastrophic forgetting and improves generation quality as indicated by metrics such as Fréchet Inception Distance (FID), distribution precision and recall, or accuracy for the downstream classification task. Ablation studies confirm the critical role of each component.