This work introduces Page-to-Token Policy Optimization (P2TPO), a structure-aligned credit-assignment framework that operates on full-page policy trajectories that outperforms all evaluated policy optimization baselines in seed-averaged scores on each of the six document parsing benchmarks.
Yong-Zhou-Shi-、-Lan-Shan-Xian-Ti-Gai-Wei-Lian-He-D Zu, Ning-Hua Yang, Ze-Zhong Tan et al.· 0 citations
Conditional memory expands language-model capacity through selective lookup, but the full embedding table must remain available for every possible input. We introduce Conditional Program (CP), a compressed representation that reconstructs only the rows an input requests. CP represents each row by a short code and a loc...
Donghoon Han, Minseop Jung, Dongjin Kwak et al.· 0 citations
Experiments with pretrained diffusion language models show that superposition-based generation can retain information from conditioning components not selected at the final measurement, while QIR improves LLM-assessed text quality when applied at an appropriate stage of generation.
A. Tanaka, Y. Akamatsu, Takuki Tomita et al.· 0 citations
Policy-shift distillation reuses the difference between a teacher's log probabilities before and after reinforcement learning (RL) as supervision on student-generated trajectories. However, this endpoint policy shift cannot reveal whether local action preferences evolved consistently during teacher training. The same e...
Bingxi Hou, Guochao Jiang, Guofeng Quan et al.· 0 citations
Reach audiences
Advertise in front of researchers, engineers, and readers.
We study sparse and continual reinforcement learning fine-tuning of language models through attention-head ranking and partitioning. For sparse learning, we rank attention heads by gradient norm to identify a small subset of heads that can be trained effectively. For continual learning, we partition heads across tasks...
Rui-Zhe Shi, Zai-Wen Yang, Qiu-Bo Wang et al.· 0 citations
Consumer phones and wearables increasingly collect physiological time-series data, yet most mobile systems treat these streams as summaries or text rather than as temporal evidence. This representation is poorly matched to personal health systems that need to retrieve, route, or summarize sensor evidence while preservi...
Max Rosenblattl, Nikhil Krishnaswamy, Leon Nissen et al.· 0 citations
Time-series data generated continuously across industrial, environmental, and especially health domains produce multimodal signals whose scale and heterogeneity make their processing and interpretation particularly challenging. Time-Series Language Models (TSLMs) promise scalable interpretation of these signals, yet re...
Max Rosenblattl, Milana Gurbanova, Karl Aurel Deck et al.· 0 citations
Language models are often asked to make judgments: whether a proposed answer is correct, whether one statement follows logically from another, or whether a response is safe. Yet changing the instructions that specify how the judgment should be expressed can make the final answer less reliable, even for the same input....
Chaemin Jang, Donggyu Lee, Jeesoo Kim et al.· 0 citations
INCONVENIENT is introduced, a benchmark designed to study robustness to convenience-sample bias and provides explicit control over both the direction and severity of relationship shifts, enabling a systematic assessment of robustness to convenience sampling.
Sara Rezaeimanesh, C. Klochko, Kundan Thind et al.· 0 citations
Dementia affects millions of older adults worldwide, yet early detection remains difficult outside clinical settings. Traditional models primarily rely on text or voice-based responses, capturing only a limited view of human cognition despite the inherently multimodal nature of human expression. This work proposes a co...
Sankalp Vaish, Rishank Singh, Sumukh Veeramalla et al.· 0 citations
Early detection of mild cognitive impairment remains a critical bottleneck in dementia care, demanding scalable tools capable of capturing subtle behavioral decay. To capture the subtle temporal changes of cognitive decline, we introduce a behavioral dataset collected across seven unscripted multimodal interactions ove...
Sankalp Vaish, Sumukh Veeramalla, Rishank Singh et al.· 0 citations
Reinforcement learning with verifiable rewards often trains language models on responses generated by a different, possibly stale behavior policy. Correcting this mismatch with learner-to-behavior importance weights can introduce high variance for long responses. We introduce Gibbs-Gap Policy Optimization (GGPO), which...
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.
