Risultati della ricerca - (coding OR Decoding) algorithm

Language Models that Think, Chat Better di Bhaskar, Adithya, Ye, Xi, Chen, Danqi

Sommario: “... rule-based rewards in verifiable domains such as mathematics and code. However, RLVR leads to limited...”
Testo

Active Attacks: Red-teaming LLMs via Adaptive Environments di Yun, Taeyoung, St-Charles, Pierre-Luc, Park, Jinkyoo, Bengio, Yoshua, Kim, Minsu

Sommario: “... exploration, we introduce \textit{Active Attacks}, a novel RL-based red-teaming algorithm that adapts its...”
Testo

From Static to Dynamic: Adaptive Monte Carlo Search for Mathematical Process Supervision di Ma, Jie, Qi, Shihao, Xing, Rui, Yin, Ziang, Wei, Bifan, Liu, Jun, Liu, Tongliang

Sommario: “..., it enhances the path expansion through a Monte Carlo algorithm with a temporally adaptive policy that begins...”
Testo

Vision-Zero: Scalable VLM Self-Improvement via Strategic Gamified Self-Play di Wang, Qinsi, Liu, Bo, Zhou, Tianyi, Shi, Jing, Lin, Yueqian, Chen, Yiran, Li, Hai Helen, Wan, Kun, Zhao, Wentian

Sommario: “...-SPO), a novel training algorithm that alternates between Self-Play and reinforcement learning...”
Testo

Who Should Have a Place on the Ark? Parameterized Algorithms for the Maximization of Phylogenetic Diversity di Schestag, Jannik

Sommario: “... after which they can no longer be saved.For Time-PD,we present color-coding algorithms that prove...”
Testo

FOLK: Fast Open-Vocabulary 3D Instance Segmentation via Label-guided Knowledge Distillation di Wu, Hongrui, Gao, Zhicheng, Cao, Jin, Yao, Kelu, Shen, Wen, Wei, Zhihua

Sommario: “... propose a label-guided distillation algorithm to distill open-vocabulary knowledge from label-consistent 2...”
Testo

Probabilistic Hyper-Graphs using Multiple Randomly Masked Autoencoders for Semi-supervised Multi-modal Multi-task Learning di Mihai-Cristian, Pîrvu, Leordeanu, Marius

Sommario: “... the distribution of hyper-edges on each forward pass. Additionally, the model adapts the standard MAE algorithm...”
Testo

FedAdamW: A Communication-Efficient Optimizer with Convergence and Generalization Guarantees for Federated Large Models di Liu, Junkang, Shang, Fanhua, Liu, Hongying, Tian, Yuxuan, Liu, Yuanyuan, Liu, Jin, Zhu, Kewen, Lin, Zhouchen

Sommario: “... \underline{AdamW} algorithm, called \texttt{FedAdamW}, for training and fine-tuning various large models...”
Testo

CORGI: Efficient Pattern Matching With Quadratic Guarantees di Weitekamp, Daniel

Sommario: “... automatically generate rules from example-driven induction or code synthesis, they can easily produce worst-case...”
Testo

MoDES: Accelerating Mixture-of-Experts Multimodal Large Language Models via Dynamic Expert Skipping di Huang, Yushi, Wang, Zining, Yuan, Zhihang, Ding, Yifu, Gong, Ruihao, Guo, Jinyang, Liu, Xianglong, Zhang, Jun

Sommario: “... the optimal thresholds, we introduce a frontier search algorithm that exploits monotonicity properties...”
Testo

Solving Spatial Supersensing Without Spatial Supersensing di Udandarao, Vishaal, Karthik, Shyamgopal, Nath, Surabhi S., Hochlehnert, Andreas, Bethge, Matthias, Prabhu, Ameya

Sommario: “... object-count predictions unchanged; instead, Cambrian-S inference algorithm relies largely on a shortcut...”
Testo

Spira: Exploiting Voxel Data Structural Properties for Efficient Sparse Convolution in Point Cloud Networks di Adamopoulos, Dionysios, Poulopoulou, Anastasia, Goumas, Georgios, Giannoula, Christina

Sommario: “... for GPUs. Spira proposes (i) a high-performance one-shot search algorithm that builds the kernel map...”
Testo

MICCAI STS 2024 Challenge: Semi-Supervised Instance-Level Tooth Segmentation in Panoramic X-ray and CBCT Images di Wang, Yaqi, Li, Zhi, Wu, Chengyu, Liu, Jun, Zhang, Yifan, Ni, Jiaxue, Luo, Qian, Chen, Jialuo, Zhang, Hongyuan, Liu, Jin, Han, Can, Fu, Kaiwen, Ji, Changkai, Cai, Xinxu, Hao, Jing, Zheng, Zhihao, Xu, Shi, Chen, Junqiang, Zhang, Qianni, Qian, Dahong, Wang, Shuai, Zhou, Huiyu

Sommario: “...) registered teams. To ensure algorithmic excellence and full transparency, we rigorously evaluated the valid...”
Testo

Semantic Soft Bootstrapping: Long Context Reasoning in LLMs without Reinforcement Learning di Mitra, Purbesh, Ulukus, Sennur

Sommario: “... optimization (GRPO), which is a commonly used RLVR algorithm. Our code is available at https://github.com...”
Testo

How Far are Modern Trackers from UAV-Anti-UAV? A Million-Scale Benchmark and New Baseline di Zhang, Chunhui, Liu, Li, Zhang, Zhipeng, Wang, Yong, Wen, Hao, Zhou, Xi, Ge, Shiming, Wang, Yanfeng

Sommario: “... of 50 modern deep tracking algorithms demonstrates that there is still significant room for improvement...”
Testo

TraPO: A Semi-Supervised Reinforcement Learning Framework for Boosting LLM Reasoning di Yang, Shenzhi, Zhu, Guangcheng, Zheng, Xing, MA, Yingfan, Chen, Zhongqi, Song, Bowen, Wang, Weiqiang, Zhao, Junbo, Chen, Gang, Wang, Haobo

Sommario: “... an effective policy optimization algorithm, TraPO, that identifies reliable unlabeled samples by matching...”
Testo

HERO-Sign: Hierarchical Tuning and Efficient Compiler-Time GPU Optimizations for SPHINCS+ Signature Generation di Zhou, Yaoyun, Wang, Qian

Sommario: “... algorithm that adapts fusion schemes to different GPU architectures. To further improve performance, HERO...”
Testo

SP-Rank: A Dataset for Ranked Preferences with Secondary Information di Hosseini, Hadi, Mandal, Debmalya, Puhan, Amrit

Sommario: “... algorithms that leverage both first-order preferences and second-order predictions in ranking tasks. Each...”
Testo

TowerMind: A Tower Defence Game Learning Environment and Benchmark for LLM as Agents di Wang, Dawei, Zhou, Chengming, Zhao, Di, Liu, Xinyuan, Ma, Marci Chi, Ushaw, Gary, Davison, Richard

Sommario: “... evaluate two classic reinforcement learning algorithms: Ape-X DQN and PPO. By offering a lightweight...”
Testo

Streaming Operator Inference for Model Reduction of Large-Scale Dynamical Systems di Koike, Tomoki, Mohan, Prakash, de Frahan, Marc T. Henry, Bessac, Julie, Qian, Elizabeth

Sommario: “... access to the underlying code or full model operators, i.e., non-intrusively. Traditional OpInf operates...”
Testo