Tracing Generated Samples to Training-Data Clusters in Flow-Matching Models
Read the original on arXiv Machine Learning →The Flow has not summarised this story yet — read it at arXiv Machine Learning.
The Flow has not summarised this story yet — read it at arXiv Machine Learning.
arXiv:2601. 22651v2 Announce Type: replace-cross Abstract: Training-data attribution for vision generative models aims to identify which training data influenced a given output.
The paper introduces CADT, a Concept Attribution method through Dynamic Trajectories, which attributes diffusion model outputs by tracking factor-induced internal response trajectories rather than scalar responses. CADT constructs matched counterfactual pairs at identical noisy states to isolate factor-specific representation displacements and models their directional and magnitude evolution across denoising as dynamic attribution signatures. Experiments on multiple public datasets demonstrate consistent improvements over existing diffusion attribution baselines for hierarchical, compositional, and style attribution tasks.
The paper introduces a new method for training data attribution in diffusion models called TID, which uses a local score discrepancy measure and can be estimated without retraining. It further distills this approach into TIDE, a forward‑only student that reproduces the teacher’s rankings using internal activations, achieving comparable accuracy at dramatically lower query cost. Experiments on CIFAR‑10, ArtBench‑10, and MS‑COCO show that TID outperforms existing methods and TIDE attributes samples in milliseconds, faster than generation itself.
FlowAlign is a flow‑guided target‑space alignment method that uses conditional flow matching as a training‑time signal to map source modality representations into a fixed target space. It preserves both the global structure of the target distribution and instance‑level correspondence in paired source–target data, unlike existing contrastive, distribution‑matching, or pointwise regression objectives. The authors provide theoretical analysis and demonstrate improved instance‑level alignment while maintaining target‑space structure on synthetic data, single‑cell multiome benchmarks, and image–text alignment tasks.
The paper introduces Differential Vector Erasure (DVE), a training‑free method for removing unwanted concepts from flow matching image generation models. DVE exploits the directional structure of the velocity field that encodes semantic concepts, constructing a differential vector field to isolate and suppress target concepts during inference. Experiments on the FLUX model show that DVE outperforms existing baselines in tasks such as NSFW suppression, style removal, and object erasure while maintaining image quality and diversity.
arXiv:2611.17767v1 Announce Type: cross Abstract: Flow-based generative models are trained on a hand-chosen prediction target, such as the velocity, the noise, or the clean data. This choice strongly...