CMU Machine Learning

Even advanced LLMs struggle to replicate human researchers' ability to find inspiring prior work, highlighting a significant gap in AI's scientific intuition.

Several open research questions are identified, including how to efficiently record incidents and how to determine whether vulnerabilities and incidents generalize, and privacy requirements are summarized and research directions for the secure and trustworthy deployment of AI agents are outlined.

ModAR is introduced, the first WAM to autoregressively denoise multiple future modalities before predicting actions, and it is found that ModAR's sequential generation outperforms existing WAM formulations, with the highest average success rate at all evaluated data scales.

LACE (Layer-Adaptive Codec Encoding), a dynamic frame rate codec that applies an independent compression step at each quantization layer, enabling layer-specific segmentation boundaries, is proposed.

This work formalizes poison selection as oracle-budgeted set optimization and introduces SAILS (Set-level Audit-Informed Iterative Learned Selection), which learns a set scorer from a few hundred finetune-and-evaluate runs, ranks millions of candidate sets, and audits only a small shortlist.

Modern VLMs already encode much of what is needed for specialized domains; the challenge is learning how to ask, suggests modern VLMs already encode much of what is needed for specialized domains; the challenge is learning how to ask.

DCP provides a common evidence language for useful outcomes, alternative routes, and feedback effects across AI research, and is applied to Core, recovered, and audit-incomplete decisions.

Retrieval harm can significantly impact performance, and simple baselines often outperform complex models in Active RAG systems.

Nearly optimal dimension lower bounds reveal that single-vector embeddings for MAX-IP require significantly more dimensions than previously thought, closing a critical gap in the literature.

RecRec reveals that decoupling reasoning from prediction can significantly enhance sequential recommendation performance, breaking free from the constraints of fixed-dimensional states.

TALRanker autonomously navigates the trade-off between efficiency and accuracy in document reranking, achieving state-of-the-art results without the latency of excessive tool calls.

Prompting decoder-only models outperforms fine-tuning methods, achieving unprecedented effectiveness in ranking case-law sentences for statutory term retrieval.

Generative AI agents can reveal how personalization algorithms amplify toxic content in ways that vary dramatically by user ideology.

Delayed start behavior can predict standardized test performance, revealing critical insights into student motivation and engagement.

MPE achieves superior long-context retrieval by efficiently encoding document chunks while maintaining critical contextual relationships, outperforming traditional methods.

Agents can now share and reuse knowledge, cutting down task execution time and improving performance without the need for coordination or joint training.

Efficient evidence ordering can cut response times by up to a third without sacrificing answer quality in RAG systems.

Students with lower self-efficacy can achieve greater learning gains when they favor the tutoring method, challenging assumptions about the superiority of technology in education.

Diminishing returns in parallel sampling can be overcome by generating diverse initial queries, leading to substantial performance gains in multi-hop question answering.

YouTube's recommendation algorithm pushes Kyrgyz children towards Russian-language content, even when they signal a preference for their native tongue, effectively amplifying colonial influence.

Untangling evidence validation from text generation, ArbGraph offers a way to build more reliable long-form RAG systems by explicitly resolving factual conflicts *before* generation even begins.

Current user modeling benchmarks are child's play compared to the real-world challenges exposed by HORIZON, a massive new dataset spanning 54M users and diverse domains.

Dramatically improve multimodal recommendation accuracy without any training by initializing user embeddings with item modality features and user cluster information.

Stop evaluating AI systems in isolation: marketplace dynamics like user switching and early-adoption advantages critically shape real-world success.

LLMs can now tap into arbitrarily long-term memories by retrieving "thoughts" – their own past reasoning steps – rather than just raw data, leading to significant performance gains.

Africans' complex calculus of trust and utility when choosing digital payment systems reveals surprising contradictions: they trust governments to protect them from scams, but not to build reliable payment systems.

Forget full retraining: intelligently selecting data subsets using gradient-based representations can keep your generative recommender fresh and robust to drift.

Forget what you know: RAG's marginal utility hinges on model scale, task type, and pretraining saturation, offering a quantitative guide to balancing pretraining and retrieval.

Forget specialized tools: a standard Unix terminal and clever RL are all you need to beat much larger LLMs at code search.

Achieving up to 3.2x speedup in recommender system inference across multiple hardware platforms without manual tuning could revolutionize deployment strategies in heterogeneous environments.