aiqbench analysis
This arXiv preprint introduces LoCo, a model-agnostic method for locating where and when visual concepts emerge across the layers, scales, and positions of next-scale autoregressive image and video models. The authors report that the localized positions are causally linked to concept emergence and use them for concept erasure, personalization, and adversarial concept injection. These are academic results on benchmarks and a new dataset, not production-ready tools, so readers should check the reported gains against independent replication.
Localizing Concepts in Visual Autoregressive Models
arXiv:2610.40061v1 Announce Type: cross Abstract: Understanding the internal routing of visual concepts is crucial for the safe and controllable adaptation of generative models. While concept localization has been widely studied in diffusion models, the emerging paradigm of next-scale visual autoregressive models remains largely unexplored. In this paper, we introduce LoCo, the first model-agnostic framework to localize where and when specific concepts emerge within autoregressive models. Driven by the native coarse-to-fine nature of next-scale generation, our method precisely maps conceptual knowledge across three distinct dimensions: Layer, Scale, and Position. To systematically localize and evaluate concept routing without context bias, we propose LoCoBench, a comprehensive dataset spanning 10 diverse categories. Extensive probing on image and video autoregressive models like Infinity, HunyuanImage-3.0, and InfinityStar shows that the localized positions are both interpretable and causally related to concept emergence. Building on these insights, we apply our localization method to three key applications: concept erasure, model personalization, and adversarial concept injection. Experiments demonstrate that our targeted intervention achieves state-of-the-art performance, substantially reducing computational overhead while preserving benign utility. Overall, our findings offer insights into how conceptual knowledge is routed during autoregressive generation, introducing a practical pathway for more interpretable, efficient, and secure adaptation.
Abstract:Understanding the internal routing of visual concepts is crucial for the safe and controllable adaptation of generative models. While concept localization has been widely studied in diffusion models, the emerging paradigm of next-scale visual autoregressive models remains largely unexplored. In this paper, we introduce LoCo, the first model-agnostic framework to localize where and when specific concepts emerge within autoregressive models. Driven by the native coarse-to-fine nature of next-scale generation, our method precisely maps conceptual knowledge across three distinct dimensions: Layer, Scale, and Position. To systematically localize and evaluate concept routing without context bias, we propose LoCoBench, a comprehensive dataset spanning 10 diverse categories. Extensive probing on image and video autoregressive models like Infinity, HunyuanImage-3.0, and InfinityStar shows that the localized positions are both interpretable and causally related to concept emergence. Building on these insights, we apply our localization method to three key applications: concept erasure, model personalization, and adversarial concept injection. Experiments demonstrate that our targeted intervention achieves state-of-the-art performance, substantially reducing computational overhead while preserving benign utility. Overall, our findings offer insights into how conceptual knowledge is routed during autoregressive generation, introducing a practical pathway for more interpretable, efficient, and secure adaptation.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
arXiv:2610.40061 [cs.CV]
(or arXiv:2610.40061v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2610.40061 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 291 words
Related coverage
CLNO: Conservative Local Neural Operator for Compressible Flow Prediction
arXiv · Oct 14, 2026
Transformer Field Theory: A Response-Theoretic Approach to Mechanistic Interpretability
arXiv · Jun 12, 2026
When Confidence Signals Disagree: Local and Global Confidence in Autoregressive Language Models
arXiv · Sep 16, 2026
Multivariate Scientific Data Compression with Learned Cross-Variable Latent Decorrelation and Autoregressive Entropy Modeling
arXiv · Sep 1, 2026