CMU Machine Learning

LightCrafter achieves superior video relighting by integrating PBR with diffusion models, enabling intricate lighting control and long-form temporal consistency without the need for extensive training data.

Agents can now work independently on data changes while humans maintain oversight, revolutionizing collaborative data management.

Every program in the new Calf framework must preserve both abstraction and potential, revolutionizing how we approach cost verification in type theory.

Operational reframing emerges as a critical risk signal, revealing that compliance can vary significantly across models and scenarios, challenging the notion of stable safety metrics in multi-agent LLMs.

Achieving trajectory-level differential privacy in adaptive streaming contexts without sacrificing performance is now feasible through an auditable buffering-aggregation approach.

Achieving up to 98.75% accuracy in detecting autism-related behaviors highlights the potential of sequence-based models over traditional CNNs in data-scarce environments.

Multimodal unlearning could revolutionize how we handle sensitive data in AI, enabling targeted removal without sacrificing model performance.

ExplAIner can express a diverse array of explanation types while ensuring efficient evaluation, transforming how we approach interpretability in machine learning models.

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

GPT-4o can identify antisemitic incidents but requires better prompts to enhance its classification accuracy.

RABBiT can accurately predict brain responses to speech with just 10 minutes of participant data, outperforming traditional models and enabling scalable population-level studies.

None of the 30 LLM agents evaluated in CausalGame demonstrated reliable causal thinking, revealing a critical gap in AI's ability to perform scientific reasoning.

Many robotic policies that seem successful in manipulation tasks actually compromise safety, with SoftVTBench revealing a stark contrast between goal completion and physical safety metrics.

Static prompts in RL training can hinder performance, but LLM-as-a-Tutor dynamically adapts them to match policy capabilities, leading to superior outcomes.

PACE-Bench predicts agentic performance with remarkable accuracy while slashing evaluation costs to a fraction of traditional methods.

Direct thread communication in MPLMs cuts context requirements by half, revolutionizing how LLMs tackle complex reasoning tasks.

Better attribution in generative music could significantly boost creator welfare and reshape platform compensation strategies.

Fixed-point flows enable a leap in performance for language models, outperforming state-of-the-art methods in one- and few-step generation tasks.

MoVA achieves superior video-text alignment by disentangling evolving visual concepts from static textual descriptions, outperforming existing models in handling long sequences.

Traditional metrics fail to capture the true memory capabilities of LLMs, exposing a critical gap in how we assess their deployment readiness.

Robots can now autonomously learn from their failures, boosting success rates by over 17% without human intervention.

ELASTIC reduces wall-clock latency by 34% while matching the success rates of the best-performing models in real-world robot manipulation.

Personalized fine-tuning of ASR models can reduce word error rates for dysarthric speech to as low as 9.7%, transforming communication for affected individuals.

Allowing language models to explore unsafe reasoning can actually enhance their ability to discern harmful from harmless prompts, reducing over-refusal without sacrificing safety.

FARS challenges the boundaries of automated research by producing 166 papers across 67 topics, revealing both its potential and pitfalls in AI-driven science.

Even top-performing AI models struggle with PowerPoint tasks, achieving only 45% success rates despite a robust evaluation framework that rewards nuanced performance.

Steering vectors can transform how we control language models, paving the way for trustworthy AI interactions in high-stakes environments.

Estimating valid transport maps can be as hard as optimal transport, but under certain conditions, alternative maps can be learned with significantly higher accuracy.

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

HTT enables tactile learning across diverse sensors, achieving adaptability that was previously unattainable in contact-rich manipulation tasks.

Grasp datasets can revolutionize robotic dexterity, enabling significant improvements in articulated tool use performance.

ANTAP achieves near-zero vulnerability to description-based attacks, fundamentally transforming how agents are evaluated and routed in multi-agent systems.

Online imitation learning can outperform offline methods, but only when the student can effectively represent the expert—realizability is key.

Despite holding privacy certifications, developers turn to Reddit for legal advice, revealing a critical gap in professional support for navigating privacy law.

Touch is not just an add-on; it fundamentally enhances object representation, leading to dramatic improvements in physical property estimation and manipulation tasks.

EFT enables LLMs to evolve solutions across diverse optimization tasks, achieving over 10% performance gains and state-of-the-art results in challenging mathematical problems.

Data mixing, especially with instruction-heavy data, emerges as the crucial factor for optimizing VLM training, challenging traditional filtering approaches.

A new AI tool can catch 34% more mathematical errors in scientific papers, transforming the peer review landscape.

EpiKV achieves 72% accuracy on MATH-500 with a 4096-token cache, rivaling the best attention-based methods while dramatically improving inference speed.

VibeAct reveals that leveraging real-time vibro-acoustic feedback can drastically enhance robotic dexterity in contact-rich environments, outperforming traditional proprioception methods.

The ABC framework empowers researchers with the largest open-source teleoperation dataset and a complete toolkit to accelerate advancements in behavior cloning for robotic manipulation.

ReStruct enables robots to adaptively steer their behavior in real-time, achieving unprecedented levels of task success and preference alignment without retraining.

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

A million p-bits in a single programmable architecture reveals a universal tradeoff between throughput and accuracy in distributed probabilistic computing.

Current multimodal conversational models miss critical emotional cues, revealing a significant gap in their ability to engage in nuanced human-like dialogue.

Local Branch Routing enables language models to leverage contextual evidence for decision-making without the computational burden of full solution searches, leading to substantial improvements in reasoning accuracy.

PatternGSL achieves accurate garment reconstruction from images without templates, revolutionizing how we think about garment design and simulation.

LALMs can achieve superior emotional comparison accuracy with only 5% of the training data typically required by conventional methods.

Current visual world models show a dramatic decline in performance when faced with unconventional and impossible physical interactions, highlighting a critical gap in their generalization capabilities.

Achieving high-fidelity language generation with 32x fewer function evaluations could revolutionize real-time applications of language models.

Decomposing annotation tasks can significantly reduce the cognitive burden on annotators, leading to better quality outputs at lower costs.

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

Score estimation errors in vanilla diffusion models can lead to catastrophic failures in compositional generation, revealing a critical gap in current methodologies.

A staggering 39.4% of open source projects may be at risk of license noncompliance due to copy-based reuse, challenging the effectiveness of existing dependency tracking methods.

Speech synthesis can now adaptively enhance clarity and vocal effort, mimicking human responses in noisy environments.

Rule-grounded reasoning can cut average distance errors in driving VLAs by nearly half, fundamentally enhancing their decision-making transparency and reliability.

Lift4D achieves unprecedented accuracy in 4D reconstruction of dynamic objects, even in the presence of severe occlusions and complex motions.

Bagpiper-TTS can seamlessly transform natural language requests into high-quality speech across diverse applications, outperforming traditional TTS systems.

Local names boost retrieval accuracy, but models still fail to generate images that faithfully represent specific street segments.

Instruction blindness in VLA models can be mitigated by optimizing for flatter loss landscapes, leading to over 60% better adherence to language instructions.

Spec learning enables LLMs to align with user preferences using just a few instructions, outperforming traditional optimization methods without requiring costly parameter updates.

Agents should enhance causal discovery workflows without compromising the integrity of causal claims by relying solely on data and expert knowledge.

Curiosity-driven interventions in LLM tutoring can boost exploratory learning behaviors by up to 2.4x, revealing the power of language in shaping cognition.

Co-authorship with humans can significantly enhance merge rates for certain AI coding agents, but this effect vanishes when accounting for repository selection and PR structure.

Sensor placement is the key determinant of success in dexterous manipulation, with whole-hand coverage vastly outperforming fingertip-only configurations.

Reviewers approve AI-generated code more often while actually engaging less, revealing a troubling trend of habituation that could compromise code quality.

FlowDPG achieves a 92% success rate in real-world robotic manipulation, bypassing the computational pitfalls of traditional policy gradient methods.

Off-policy degree in RLVR updates can drastically change which tokens drive learning, leading to a new adaptive method that outperforms traditional baselines.

Discretizing reward models can significantly enhance policy performance by reducing oversensitivity without sacrificing discriminative ability.

Expert iteration emerges as the key driver of quality in text-to-music generation, overshadowing the contributions of preference tuning.

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

Phoenix resolves GitHub issues with 75% accuracy while ensuring safety through a multi-agent system, but still faces challenges in planner localization.

Weak audio supervision allows ReNikud to achieve superior grapheme-to-phoneme conversion for Hebrew, outperforming traditional methods that struggle with data scarcity and pronunciation accuracy.

T2I models can be effectively probed for identity memorization without any access to training data, revealing surprising differences in how they handle famous versus less recognized names.

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

S-JEPA sets a new standard in speech representation learning by achieving top performance with fewer parameters and without the cumbersome offline re-clustering process.

Pro-female bias in LLM hiring decisions persists even in non-Western contexts, with candidate names being the key driver of this bias.

Idempotency in training voice attribute editing models can drastically reduce the impact of noisy labels, leading to more reliable and consistent edits.

Machine learning resolves 20,000 ambiguous X-ray source matches, revealing the limitations of traditional spatial cross-matching methods.

Generalist agents need to remember distinct information to navigate conflicting optimal actions across environments, challenging the notion that current state observations are sufficient.

Training performance can significantly forecast real-life tutoring effectiveness, with open responses proving to be a stronger predictor than traditional assessments.

LLM agents can identify reproducibility problems in 90% of analyzed machine learning papers, leveraging GitHub issues as a novel supervision source.

Vision Transformers can dramatically reduce demographic bias in face PAD systems, achieving an 83% reduction in error gaps across ethnic groups.

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.

AIGS-Net achieves superior low-light image enhancement with only 40 learnable parameters, redefining the trade-off between quality and efficiency in image processing.

Dixtral achieves up to 29% absolute improvement in speaker-attributed transcription accuracy by leveraging diarization masks without risking catastrophic forgetting.

Commercial LLMs may seem reliable for security advice, but they can deliver contradictory responses, risking user safety and trust.

Training on compound reasoning traces yields better generalization than isolated atomic modules, reshaping our understanding of how LLMs can learn to reason.

LLMs can outperform humans in predicting the next speaker in meetings, even without audio or visual data.

Traditional success metrics tie agent performance 75% of the time—this new approach slashes that to 35%, revealing clearer distinctions in agent capabilities.

ExpRL outperforms traditional reinforcement learning methods by effectively rewarding intermediate reasoning steps, leading to better LLM performance on complex tasks.

ART-Glove captures human dexterity and contact interactions with unprecedented precision, paving the way for more effective robot learning.

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

The quality cliff at low frame rates is driven by training configuration issues, not inherent limitations of neural audio codecs.

LLMs show significant variability in the actionability of their UX critiques, with some models outperforming others across different product categories.

TuneJury achieves superior music preference alignment with a single frozen reward model that adapts efficiently to new audio generators.

$λ$-Reachability improves safety analysis in robotics by significantly enhancing the accuracy of safety margin estimations and safe-set classifications.

AIChilles reveals 49 hidden weaknesses in AI-evolved systems, challenging the assumption that AI-generated code is always superior to human-designed algorithms.

Joint image-depth generation can be achieved with a single model trained on sparse data, outperforming existing methods by a significant margin.

Reducing inter-utterance silence from 9.6 seconds to 0.3 seconds transforms the quality of real-time game commentary, making it feel more natural and engaging.