Judge Decoding accelerates LLM inference by relaxing the strict verification of Speculative Decoding, yet it typically relies on expensive and noisy supervision. In this work, we revisit this paradigm from first principles, revealing that the ``criticality'' scores learned via costly supervision are intrinsically encod...
Shengyin Sun, Yiming Li, Renxi Liu et al.· 0 citations
In this work, we examine hateful memes from three complementary angles - how to detect them, how to explain their content and how to intervene them before being posted - by applying a range of strategies built on top of generative AI models. To the best of our knowledge, explanation and intervention have typically been...
Naquee Rizwan, Subhankar Swain, Paramananda Bhaskar et al.· 4 citations
This work formalizes polarity-aware evidence grounding: linking each response sentence to its contextual evidence with a supporting, contradicting, or irrelevant label, and proposes eTracer, a post-hoc framework whose context-aware signed scorer recovers the full evidence map in a single forward pass per response.
Bo-Hao Chu, Qianli Wang, Hendrik Damm et al.· 0 citations
A human-annotated benchmark that grounds each summarized claim in both cited sentences and contributory phrases within them and introduces a decoupled evaluation framework that separately scores content, traceability, and locatability, together with a proxy for the amount of source text a clinician must inspect to veri...
Bo-Hao Chu, Hendrik Damm, T. Pakull et al.· 0 citations
Reach audiences
Advertise in front of researchers, engineers, and readers.
IndicParam, a human-curated benchmark of over 13,000 multiple-choice questions covering 11 Indic languages plus Sanskrit-English code-mixed set, provides insights into limitations of cross-lingual transfer and establishes a challenging benchmark for Indic languages.
Analysis of reasoning errors in vision language models and how they impact user trust and the ability to detect errors highlights how CoT explanations can simultaneously clarify and mislead, underscoring the need for NLP systems to provide explanations that encourage scrutiny and critical thinking rather than blind tru...
Eunkyu Park, Wesley Hanwen Deng, Vasudha Varadarajan et al.· arXiv.org· 6 citations
Parallel tokenizers are introduced, a new framework that first trains tokenizers monolingually and then aligns their vocabularies exhaustively using bilingual dictionaries or word-to-word translation to enforce a shared semantic space across languages while naturally improving fertility balance.
Muhammad Dehan Al Kautsar, Fajri Koto· 0 citations
This work proposes a new paradigm: Test-Time Policy Adaptation for Multi-Turn Interactions (T2PAM), which utilizes user feedback from the ongoing interaction as a reward signal to estimate a latent optimal policy aligned with user preferences, then updates a small subset of parameters to steer the model toward this pol...
Chen-Xing Wei, Hong Wang, Ying He et al.· arXiv.org· 4 citations
Stronger learning signals do not necessarily produce a stronger small reasoning model. Off-policy supervised fine-tuning (SFT) supplies stronger traces that may be incompatible with the student. On-policy post-training instead remains constrained by the quality of the student's own trajectories: reinforcement learning...
Jaehoon Kim, Kwangwook Seo, Dongha Lee· 0 citations
Demographic analysis across 20 Jigsaw and 13 ToxiGen minority groups reveals structurally unequal adversarial vulnerability, exposing mechanistically traceable fairness gaps in current toxicity classifiers.
Shaz Furniturewala, A. Zubiaga· arXiv.org· 1 citation
PID-steering is described, a training-free, decoding-time method that modulates the steering strength with a PID controller driven by a lightweight chunk-level redundancy classifier.
Applying ALAS to four open-source Speech-LLMs across emotion recognition, open-ended SQA, and multi-choice audio understanding, it is found that the depth and strength of alignment reflect each model's audio-encoder design and the acoustic-versus-semantic demands of the task, and that ALAS tracks but does not duplicate...
Pooneh Mousavi, Yingzhi Wang, M. Ravanelli et al.· 4 citations
Flint-interactive gives AI agents a language for creating interactive visualizations. Click a chart, filter data, highlight related content, and explore beyond the view the author originally chose. The post Flint-interactive: A language for creating interactive data visualizations appeared first on Microsoft Research.
A new computational approach reveals how cells communicate to assemble into tissues. This could accelerate stem cell engineering, improve disease models, and inform therapies.
Microsoft Research Blog· microsoft.comOct 15, 2026
Zero-knowledge proofs can verify computing results without rerunning workloads or exposing sensitive data. Magnetar helps bring them into the mainstream with support for standard programming languages, major speed gains, and quantum-resistant security. The post Proof-based general-purpose computing with Magnetar appeared first on Microsoft Research.
MIT News · Artificial Intelligence· news.mit.eduSep 24, 2026
