Research Blogs News Tags Projects

Research

Select a Tag

Published on
June 5, 2025
Uncertainty Quantification and Confidence Calibration in Large Language Models: A Survey
high-stakes-domains uncertainty-quantification confidence-calibration taxonomy-of-UQ-methods reliability open-challenges decision-making real-world-applicability reasoning input-ambiguity reasoning-path-divergence uncertainty-dimensions interpretable-UQ-approaches computational-constraints robust-UQ-approaches scalable-UQ-approaches decoding-inconsistencies decoding-stochasticity text-generation risk-mitigation Large-Language-Models computational-efficiency
• Arizona State University, Tempe, Arizona, USA
Published on
June 5, 2025
LSC-Eval: A General Framework to Evaluate Methods for Assessing Dimensions of Lexical Semantic Change Using LLM-Generated Synthetic Data
Lexical-Semantic-Change synthetic-datasets Sentiment In-Context-Learning Intensity social-sciences computational-methods evaluation-framework Breadth SIBling-framework
• Melbourne School of Psychological Sciences, The University of Melbourne• School of Computing and Information Systems, The University of Melbourne• School of Electronic Engineering and Computer Science, Queen Mary University of London
Published on
June 5, 2025
Escaping Plato’s Cave: Towards the Alignment of 3D and Text Latent Spaces
retrieval-tasks lower-dimensional-subspaces 3D-encoders uni-modal-representations geometric-data-representations semantic-data-representations matching-tasks text-encoders feature-alignment
• École polytechnique• Sapienza University of Rome• The University of Texas at Austin• University of Milano-Bicocca
Published on
June 5, 2025
Knowledge Retention for Continual Model-Based Reinforcement Learning
continual-model-based-reinforcement-learning learning-and-adaptation Synthetic-Experience-Rehearsal world-models empirical-evaluations generative-models knowledge-retention intrinsic-reward-mechanism reward-functions
• Department of Artificial Intelligence, University of ABC• Department of Computer Science, University of XYZ• Department of Computer Science, University of DEF• Department of Robotics, University of GHI
Published on
June 5, 2025
The Role of Visual Modality in Multimodal Mathematical Reasoning: Challenges and Insights
model-performance image-encoders visual-perception-capabilities multimodal-mathematical-reasoning visual-question-answering-(VQA)datasets benchmarks visual-information
• School of Computer Science and Technology, East China Normal University• Meituan Inc.• Fudan University
Published on
June 5, 2025
Dynamic Benchmarking of Reasoning Capabilities in Code Large Language Models Under Data Contamination
code-large-language-models dynamic-data-generation empirical-studies benchmarking data-contamination reasoning-capabilities
• Department of Computer Science, University of XYZ• Department of Electrical Engineering, University of ABC• Department of Mathematics, University of DEF
Published on
June 5, 2025
On the Acquisition of Shared Grammatical Representations in Bilingual Language Models
bilingual-models language-pairs grammatical-representations multilingual-capabilities crosslingual-transfer structural-priming language-exposure data-control asymmetrical-effects transfer-learning typologically-diverse-languages
Published on
June 5, 2025
InSerter: Speech Instruction Following with Unsupervised Interleaved Pre-training
text-to-speech-conversion SOTA-performance speech-large-language-models representation-alignment semantic-inconsistency behavior-alignment speech-instructions unsupervised-pre-training SpeechInstructBench speech-instruction-following-tasks
• The Chinese University of Hong Kong• Alibaba Group
Published on
June 5, 2025
Where Are We? Evaluating LLM Performance on African Languages
NLP policy-reforms linguistic-diversity empirical-evaluation African-languages inclusive-data-practices data-inequities large-language-models
• The University of British Columbia• MBZUAI
Published on
June 5, 2025
Sliding Window Attention Training for Efficient Large Language Models
transformer-based-Large-Language-Models attention-sink-phenomenon sparse-attention ALiBi SOTA-performance Rotary-Position-Embedding efficiency-of-LLMs state-space-models computational-complexity linear-recurrent-architectures softmax-operation Sliding-Window-Attention-Training information-compression sigmoid-function
• City University of Hong Kong• Xi’an Jiaotong University• Jarvis Research Center, Tencent YouTu Lab• Jarvis Research Center, Tencent YouTu Lab, Westlake University
Published on
June 5, 2025
VCT: Training Consistency Models with Variational Noise Coupling
CIFAR-10 diffusion-models image-generation forward-kernels Variational-Consistency-Training training-variance ImageNet Variational-Autoencoders learned-noise-data-coupling training-dynamics flow-matching Consistency-Training
• University of XYZ• Institute of ABC• National University of DEF• Research Center GHI
Published on
June 5, 2025
Aligning Compound AI Systems via System-level DPO
SysDPO-Sampling language-model external-tools SysDPO SysDPO-Direct non-differentiable-interactions Directed-Acyclic-Graphs LLM-collaboration-system component-level-preferences diffusion-model policy-optimization Direct-Preference-Optimization foundation-models Compound-AI-systems end-to-end-gradient-based-optimization LLMs system-level-preferences
• Stanford University• Mila - Quebec AI Institute
Published on
June 5, 2025
FlexTok: Resampling Images into 1D Token Sequences of Flexible Length
GPT-style-Transformer 1D-tokenization rectified-flow-model autoregressive-image-generation variable-length-token-sequences compressed-representations image-tokenization text-conditioned-image-generation
Published on
June 5, 2025
Why Safeguarded Ships Run Aground? Aligned Large Language Models’ Safety Mechanisms Tend to Be Anchored in The Template Region
vulnerabilities safety-alignment mechanistic-analyses template-region safety-mechanisms jailbreak-attacks large-language-models
• Department of Computing, The Hong Kong Polytechnic University• Zhejiang University
Published on
June 5, 2025
Refining Sentence Embedding Model through Ranking Sentences Generation with Large Language Models
SOTA-performance sentence-embedding contrastive-learning latent-space annotation-dependency ranking-information semantic-distinctions large-language-models
• State Key Laboratory of Cognitive Intelligence, University of Science and Technology of China• State Key Laboratory of Cognitive Intelligence, University of Science and Technology of China; Institute of Artificial Intelligence, Hefei Comprehensive National Science Center• Shenzhen International Graduate School, Tsinghua University• Zhejiang University
Published on
June 5, 2025
Continual Release Moment Estimation with Differential Privacy
Gaussian-density-estimation matrix-mechanism sensitivity-analysis Joint-Moment-Estimation DP-Adam differential-privacy CIFAR-10
• Institute of Science and Technology Austria (ISTA), Klosterneuburg, Austria• Department of Computer Science, Rutgers University, Piscataway, NJ 08854, USA
Published on
June 5, 2025
PiKE: Adaptive Data Mixing for Large-Scale Multi-Task Learning Under Low Gradient Conflicts
data-mixing language-model-pretraining gradient-conflicts multi-task-learning theoretical-convergence-guarantees adaptive-algorithms
• University of Southern California• Google Research
Published on
June 5, 2025
Theoretical Analysis of KL-regularized RLHF with Multiple Reference Models
multiple-reference-models KL-regularization human-feedback theoretical-framework sample-complexity reinforcement-learning large-language-models AI-alignment-techniques
Published on
June 5, 2025
MM-IQ: Benchmarking Human-Like Abstraction and Reasoning in Multimodal Models
IQ-testing reasoning-paradigms abstraction reasoning reward-functions reinforcement-learning visual-reasoning-problems evaluation-framework human-cognitive-capabilities multimodal-models multimodal-systems
• Tencent Hunyuan Team
Published on
June 5, 2025
OneForecast: A Universal Framework for Global and Regional Weather Forecasting
deep-learning-models dynamic-system-modeling graph-neural-networks multi-grid-theory extreme-event-predictions weather-forecasting high-resolution-forecasts numerical-weather-prediction
• University of Science and Technology• Tsinghua University• Peking University• Zhejiang University
Published on
June 5, 2025
Neural Discovery in Mathematics: Do Machines Dream of Colored Planes?
extremal-combinatorics discrete-geometry optimization-task neural-networks six-colorings gradient-based-exploration mathematical-discovery Hadwiger-Nelson-problem coloring-the-plane probabilistic-loss-function
Published on
June 5, 2025
Through the Prism of Culture: Evaluating LLMs’ Understanding of Indian Subcultures and Traditions
cultural-sensitivity Little-Traditions cultural-bias AI-systems Great-Traditions Indian-subcultures Large-Language-Models response-quality prompting-strategies
Published on
June 5, 2025
Weight for Robustness: A Comprehensive Approach towards Optimal Fault-Tolerant Asynchronous ML
optimal-convergence-rate weighted-robust-aggregation fault-tolerance Byzantine-robust-training variance-reduction-technique asynchronous-distributed-machine-learning
• Department of Electrical Engineering, Technion, Haifa, Israel
Published on
June 5, 2025
ExeChecker: Where Did I Go Wrong?
rehabilitation-exercises graph-attention-neural-networks contrastive-learning human-pose-estimation pairwise-sequence-alignment transformer-interpretability in-house-dataset informative-feedback
• Department of Computer Science, University of Massachusetts Boston• Department of Electrical and Computer Engineering, University of Massachusetts Amherst• Department of Computer Science, Boston University
Published on
June 5, 2025
MAmmoTH-VL: Eliciting Multimodal Reasoning with Instruction Tuning at Scale
CoT-reasoning instruction-response-pairs dataset-construction benchmark-performance multimodal-large-language-models reasoning-capabilities instruction-tuning
• M-A-P• Nanyang Technological University• University of Waterloo• The University of Manchester
Published on
June 5, 2025
Learning 3D Representations from Procedural 3D Programs
shape-classification 3D-shapes part-segmentation procedural-3D-programs self-supervised-learning masked-point-cloud-completion 3D-representations unlabeled-3D-point-clouds
• University of Virginia
Published on
June 5, 2025
KVPR: Efficient LLM Inference with I/O-Aware KV Cache Partial Recomputation
data-movement recomputation scheduling Large-Language-Models auto-regressive-decoding GPU-memory KV-cache inference-performance PCIe-bandwidth profiling execution-plan heterogeneous-execution CPU-memory
• University of Southern California
Published on
June 5, 2025
Sonic: Shifting Focus to Global Audio Perception in Portrait Animation
audio-driven-animation time-aware-position-shift-fusion motion-diversity temporal-consistency context-enhanced-audio-learning talking-face-generation facial-movements video-quality lip-synchronization-precision motion-decoupled-controller global-audio-perception
• Tencent• Zhejiang University
Published on
June 5, 2025
Large-Scale Text-to-Image Model with Inpainting is a Zero-Shot Subject-Driven Image Generator
subject-alignment inpainting text-to-image-generation image-prompting diptych-generation image-generation-applications zero-shot-approaches attention-weights
• Data Science and AI Laboratory, ECE, Seoul National University• AIIS, ASRI, INMC, ISRC, and Interdisciplinary Program in AI, Seoul National University
Published on
June 5, 2025
VinePPO: Refining Credit Assignment in RL Training of LLMs
reasoning-tasks MATH-dataset GSM8K-dataset credit-assignment reinforcement-learning Monte-Carlo-based-estimates large-language-models value-networks Proximal-Policy-Optimization
• University of Tehran• Sharif University of Technology• University of Montreal• Mila - Quebec AI Institute