Large foundation models often answer even when the available visual or textual evidence is insufficient to support a reliable conclusion. This failure to abstain makes them hard to trust in high-stakes or information-seeking settings, where a confident but unsupported answer can be worse than no answer at all. It also raises a natural question: can the internal mechanisms behind this behaviour be identified and controlled at inference time? We present Dr. No, a feature-steering approach that mines sparse autoencoder (SAE) features associated with abstention and commitment, then modulates a small set of them during the prefill pass, without retraining the base model. Dr. No is evaluated on LLaVA-1.6 Mistral 7B across multimodal abstention benchmarks, including UNK-VQA and MM-UPD, and additionally benchmarked on language reasoning tasks using Qwen3.5-2B on AbstentionBench subsets. Against multiple uncertainty-based abstention baselines, feature steering achieves a more favourable balance between correct answering and appropriate abstention than post-hoc rejection methods. The gains are strongest on UNK-VQA and on MM-UPD cases involving image--question or question--answer incompatibility, and transfer most clearly to mathematical reasoning subsets of AbstentionBench. However, performance is more limited on absent-answer detection and retrieval-heavy multi-hop QA, such as MuSiQue. These results suggest that sparse internal features provide lightweight control handles for shaping uncertainty-aware behaviour in foundation models, while also showing that abstention control is source-dependent rather than governed by a single universal signal.
Francesco Di Sario, Riccardo Renzulli, Marco Grangetto, Akihiro Sugimoto, Enzo Tartaglione
原文摘要
Splat-based radiance fields have rapidly diversified in primitive type, kernel support, and appearance parameterization. This diversity makes compression difficult: most codecs are single-rate, requiring a separately optimized model for each target bitrate, while progressive codecs remain tied to representation-specific assumptions that do not transfer across backbones. We introduce GoDe (Gaussians on Demand), a progressive compression framework whose central contribution is backbone agnosticism. It reorganizes any trained differentiable splat representation into a single scalable bitstream without modifying the backbone or re-deriving the compression structure for each new primitive type. GoDe builds a nested level-of-detail hierarchy via masked-gradient importance ranking, refines all levels with one quantization-aware fine-tuning stage, and stores the base layer and enhancement layers as progressive chunks compressed with shared entropy dictionaries. Any decoded prefix is directly renderable, so one training pipeline provides multiple rate-distortion operating points without per-rate retraining. Across five heterogeneous backbones and three standard benchmarks, GoDe achieves competitive rate-distortion performance while enabling faster rendering and substantially faster progressive decoding than existing methods. Project page: https://gaussians-on-demand.github.io/