Beyond the Veil of Similarity: Quantifying Semantic Continuity in Explainable AI

We introduce a novel metric for measuring semantic continuity in Explainable AI methods and machine learning models. We posit that for models to be truly interpretable and trustworthy, similar inputs should yield similar explanations, reflecting a consistent semantic understanding. By leveraging XAI techniques, we assess semantic continuity in the task of image recognition. We conduct experiments to observe how incremental changes in input affect the explanations provided by different XAI methods. Through this approach, we aim to evaluate the models' capability to generalize and abstract semantic concepts accurately and to evaluate different XAI methods in correctly capturing the model behaviour. This paper contributes to the broader discourse on AI interpretability by proposing a quantitative measure for semantic continuity for XAI methods, offering insights into the models' and explainers' internal reasoning processes, and promoting more reliable and transparent AI systems.

The Tree of Diffusion Life: Evolutionary Embeddings to Understand the Generation Process of Diffusion Models

Diffusion models generate high-quality samples by corrupting data with Gaussian noise and iteratively reconstructing it with deep learning, slowly transforming noisy images into refined outputs. Understanding this data evolution is important for interpretability but is complex due to its high-dimensional evolutionary nature. While traditional dimensionality reduction methods like t-distributed stochastic neighborhood embedding (t-SNE) aid in understanding high-dimensional spaces, they neglect evolutionary structure preservation. Hence, we propose Tree of Diffusion Life (TDL), a method to understand data evolution in the generative process of diffusion models. TDL samples a diffusion model's generative space via instances with varying prompts and employs image encoders to extract semantic meaning from these samples, projecting them to an intermediate space. It employs a novel evolutionary embedding algorithm that explicitly encodes the iterations while preserving the high-dimensional relations, facilitating the visualization of data evolution. This embedding leverages three metrics: a standard t-SNE loss to group semantically similar elements, a displacement loss to group elements from the same iteration step, and an instance alignment loss to align elements of the same instance across iterations. We present rectilinear and radial layouts to represent iterations, enabling comprehensive exploration. We assess various feature extractors and highlight TDL's potential with prominent diffusion models like GLIDE and Stable Diffusion with different prompt sets. TDL simplifies understanding data evolution within diffusion models, offering valuable insights into their functioning.

Unraveling the Temporal Dynamics of the Unet in Diffusion Models

Diffusion models have garnered significant attention since they can effectively learn complex multivariate Gaussian distributions, resulting in diverse, high-quality outcomes. They introduce Gaussian noise into training data and reconstruct the original data iteratively. Central to this iterative process is a single Unet, adapting across time steps to facilitate generation. Recent work revealed the presence of composition and denoising phases in this generation process, raising questions about the Unets' varying roles. Our study dives into the dynamic behavior of Unets within denoising diffusion probabilistic models (DDPM), focusing on (de)convolutional blocks and skip connections across time steps. We propose an analytical method to systematically assess the impact of time steps and core Unet components on the final output. This method eliminates components to study causal relations and investigate their influence on output changes. The main purpose is to understand the temporal dynamics and identify potential shortcuts during inference. Our findings provide valuable insights into the various generation phases during inference and shed light on the Unets' usage patterns across these phases. Leveraging these insights, we identify redundancies in GLIDE (an improved DDPM) and improve inference time by ~27% with minimal degradation in output quality. Our ultimate goal is to guide more informed optimization strategies for inference and influence new model designs.