The paper argues that the architecture of multi‑agent large language model (LLM) frameworks, rather than just the intelligence of the underlying models, largely determines system performance. It introduces a taxonomy of architectural dimensions—such as orchestration, memory, planning interfaces, specialization, and communication topology—and presents MAFBench, a unified evaluation suite. An empirical study across nine frameworks, keeping the LLM constant, reveals six design principles and shows that choices like orchestration and communication topology can dramatically affect latency, accuracy, and coordination success.
By Abdelghny Orogat, Ana Rostam, Essam Mansour
Semantic Feature Analysis (SFA) is a method that refines agent specifications without performing any rollout-based search. It analyzes existing execution traces, clusters workflow node outputs, extracts semantic feature classes via an extended subject‑verb‑object schema, ranks these features with a decision tree, and injects the most impactful features back into the system prompt. Evaluations on four benchmarks show that SFA consistently outperforms five prompt‑optimisation algorithms and a single‑reflection baseline, especially when rollout costs are high.
By Yuval David, Fabiana Fournier, Lior Limonad, Hadar Mulian
BuildBench introduces a realistic benchmark for evaluating large language model agents on the task of compiling open‑source software (OSS). It includes diverse OSS projects that lack clear build instructions, have undocumented dependencies, and may require source patching or script modification. The authors also present OSS‑BUILD‑AGENT, a baseline LLM‑based agent that retrieves build instructions effectively and achieves state‑of‑the‑art performance on the benchmark.
By Zehua Zhang, Ati Priya Bajaj, Divij Handa, Siyu Liu, Arvind S Raj, Hongkai Chen, Hulin Wang, Yibo Liu, Zion Leonahenahe Basque, Souradip Nath, Vishal Juneja, Nikhil Chapre, Tiffany Bao, Yan Shoshitaishvili, Adam Doup\'e, Chitta Baral, Ruoyu Wang
Ambient Dataloops is an iterative framework that refines datasets to improve diffusion model training. By co‑evolving the dataset and the model, each iteration produces higher‑quality data while the model learns to handle slightly less noisy synthetic samples using Ambient Diffusion techniques. The approach achieves state‑of‑the‑art results in unconditional and text‑conditional image generation as well as de novo protein design, and the authors provide a theoretical justification for the benefits of the looping procedure.
By Adri\'an Rodr\'iguez-Mu\~noz, William Daspit, Adam Klivans, Antonio Torralba, Constantinos Daskalakis, Giannis Daras
StarVLA-α is a streamlined Vision‑Language‑Action (VLA) model that reduces architectural and pipeline complexity to facilitate systematic study of VLA design choices. By employing a strong VLM backbone and minimal design, it achieves competitive performance across multiple benchmarks (LIBERO, SimplerEnv, RoboTwin, RoboCasa) and outperforms the baseline π₀.₅ by 20% on the RoboChallenge benchmark. The authors plan to release the code to support future VLA research.
By Jinhui Ye, Ning Gao, Senqiao Yang, Jinliang Zheng, Zixuan Wang, Yuxin Chen, Pengguang Chen, Yilun Chen, Shu Liu, Jiaya Jia
The paper introduces M2Tok, a Multi-head Multi-codebook Action Tokenizer that reduces reconstruction loss for continuous action signals by decomposing latent features into multiple heads and assigning independent codebooks to each. This design expands representational expressivity, leading to lower reconstruction error and higher success rates in Vision‑Language‑Action models evaluated on RoboTwin, Simpler‑Env, and zero‑shot real‑world tasks. Ablation studies confirm the effectiveness of both multi‑head and multi‑codebook mechanisms.
By Chunpu Xu, Zhixuan Liang, Yuhao Zhang, Chi-Min Chan, Jessie Wang, Yang Xiao, Mengkang Hu, Xiaokang Yang, Yao Mu
The paper introduces Stiefel Attention, which constrains the query and key projection matrices of transformers to the Stiefel manifold and optimizes them with a Riemannian Adam variant. It demonstrates that this approach yields steepest‑descent updates, is well‑conditioned, and preserves learned attention geometry during weight decay. Empirical results show significant accuracy gains on modular arithmetic grokking and CIFAR‑10 patches, with the improvement attributed to a step‑scale‑free update rule rather than equivariance or projector changes.
By Rub\'en Dar\'io Guerrero
The paper introduces Reward Stimulation Implicit Q-Learning (RSIQL), a non-hierarchical approach to improve offline goal-conditioned reinforcement learning. RSIQL adds auxiliary reward signals at intermediate states that are predicted to aid progress toward the goal, thereby reducing the delay in training supervision. Experiments on D4RL goal-reaching benchmarks and OGBench demonstrate that RSIQL outperforms baseline goal-conditioned IQL and rivals hierarchical offline methods while maintaining a simple flat policy structure.
By Jing Zhang
Graph-Based Stochastic-Power-UCT (GS-Power-UCT) is a Monte‑Carlo graph search algorithm that shares states reached at the same planning depth while keeping separate values for different depths, thereby reducing duplicate simulations in stochastic MDPs. The method guarantees that, for a fixed horizon, the root estimate converges to the finite‑horizon value at an $O(n^{-1/2})$ rate, matching tree‑based Stochastic‑Power‑UCT but with improved sample reuse. Two full‑state variants—GS‑Power‑UCT‑F and GS‑Power‑UCT‑F$^+$—extend the approach to single‑node per physical state and adaptive horizons, respectively, with the latter converging to the optimal infinite‑horizon discounted value when cross‑depth bias vanishes. Experiments on stochastic planning benchmarks demonstrate that GS‑Power‑UCT outperforms both tree‑based and other graph‑based baselines in sample efficiency.
By Tung Tran, Viet Bao Mai, Hoang Ta, Tuan Dam
The paper introduces BADA, a Boundary-Aware Data Augmentation technique for offline reinforcement learning. By interpolating neighboring states to create synthetic data that respects the original distribution, BADA improves in-distribution generalization and robustness. Experiments on limited offline datasets show that BADA achieves state-of-the-art performance across diverse benchmarks.
By Gong Gao, Weidong Zhao, Xianhui Liu
The paper introduces EviRec, a continual evidence-learning framework for dual cold-start point-of-interest (POI) recommendation. It evaluates three complementary views—matching, transition-memory, and lifecycle—to score visible POI candidates, and uses a reliability gate to adaptively combine transition-memory and lifecycle evidence based on each candidate’s observation state. Experiments on a large five-city dataset show that EviRec outperforms state-of-the-art baselines, achieving a 20.4% NDCG@10 improvement on Dual-New cases while preserving learned mobility routines.
By Rongchao Xu, Lin Jiang, Guang Wang
The paper critiques the common practice of stopword removal in legal text analysis, showing that standard stoplists actually degrade performance on binary classification tasks involving Supreme Court opinions. By exhaustively testing the removal of each of ~18,500 candidate words, the authors find that no stoplist—generic or optimized—outperforms a no‑removal baseline, and that models cannot predict which words are beneficial to remove. The study argues that inherited preprocessing defaults can distort the doctrinal and ideological signals that legal scholars aim to recover, calling into question the validity of such practices.
By Gregory M. Dickinson
EmbodiedMind introduces a three-stage training paradigm for embodied foundation models that tackles inefficient sample use, task imbalance, and credit assignment in long-horizon planning. The stages—Rejection Sampling-based Fine‑Tuning, Iterative Rejection GRPO, and Trie‑GRPO—filter low‑informative data, balance task difficulty, and use action prefix trees for step‑level advantage estimation. This approach yields a state‑of‑the‑art average performance of 70.02% across 18 benchmarks, notably improving long‑horizon task planning accuracy.
By Feifan Wang, Zongbing Zhang, Yu Zhang, Lingfeng Wang, Yurui Zhu, Jin Deng, Mingliang Zhang, Zhengguang Gao, Yongcheng Wang, Jin Xu, Ri Yang
The paper presents a signal-theoretic machine learning framework to detect deceptive online job ads that facilitate forced labour. Using 464 verified cases from nine countries and 21 industries, the authors build multimodal models that combine computer vision, natural language processing, and semantic embeddings, achieving ROC‑AUC scores between 0.87 and 0.97. SHAP analysis identifies text quality, risk language, and visual features as key discriminators, and the authors deliver a proof‑of‑concept decision support system that outputs interpretable risk scores for practitioners.
By Sajid Siraj, Mahnaz Hosseinzadeh, Amin Vafadarnikjoo, Shuyang Li
The paper presents a method for distilling a large 300‑million‑parameter geospatial foundation model (Prithvi‑EO‑2.0) into a compact 0.7‑million‑parameter EfficientViT‑B0 student for flood segmentation. By using the teacher to supervise additional unlabeled Sentinel‑2 imagery, the student’s training set expands without new manual labels, achieving competitive performance on Sen1Floods11 and STURM‑Flood while remaining smaller and faster. After quantization, the student runs as a 1.5‑MB INT8 TensorRT engine on a Jetson Xavier NX, processing 512×512 images in 5.57 ms with ~14 MB of memory.
By Fabian Schmalstieg, Karsten Mueller, Wojciech Samek
The paper introduces GenAIMMD, an iterative algorithm that learns the committor function and trains a conditioned Boltzmann Generator to generate uncorrelated transition paths without prior knowledge of the reaction coordinate. This method combines transition path sampling with committor learning, enabling fully parallelizable sampling. Benchmarks on a toy model and a polymer system show a substantial performance improvement over standard TPS.
By Maximilian Negedly, Sebastian Falkner, Alessandro Coretti, Christoph Dellago
Relational BabyLM is a decoder‑only Transformer that replaces standard self‑attention with a Dual Attention Transformer (DAT) to separate object‑level lexical features from structural/relational information. The model incorporates a Next‑Latent Prediction objective to compress history into a dense belief state and introduces a RoPE‑based symbol‑retrieval mechanism. On the BabyLM 2026 challenge, the best model ranks 6th overall and 3rd on the NLP‑task subset, outperforming GPT‑2 on most benchmarks and achieving the highest EWoK score among strict‑track entries.
By Adrian Brasoveanu, Ece Takmaz, Jakub Dotla\v{c}il
The paper introduces QUATRO, a reinforcement‑learning approach for fine‑tuning large language models that enforces trust‑region constraints directly rather than relying on heuristic clipping. By deriving a principled objective, QUATRO provides explicit control over policy updates and stabilizes entropy during training. Experiments on mathematical reasoning benchmarks demonstrate that QUATRO maintains stable training even with higher learning rates and increased policy staleness.
By Doyeon Lee, Eunyi Lyou, Hyunsoo Cho, Sookyung Kim, Joonseok Lee, Jaemoo Choi
The paper introduces Post-Boundary Bridge (PBB), a hybrid transformer architecture that keeps causal attention within blocks while adding direct connections across block boundaries. PBB focuses on within-block modeling and nearby exchange, delegating long-range communication to full-attention layers. Experiments on dense and mixture-of-experts models ranging from 205 million to 2.07 billion parameters show that PBB hybrids maintain near-full perplexity, competitive downstream performance, and improved source retrieval, while Flash-PBB achieves 1.82× faster decoding with half the local key‑value cache compared to Flash‑SWA.
By Zhibo Yang
The study evaluated nine transcriptomic models—five bulk RNA‑seq and four single‑cell RNA‑seq—designed to predict response to immune checkpoint inhibitors. Across independent datasets, bulk models performed near chance while single‑cell models offered only modest gains, and pathway analyses revealed inconsistent biomarker signals. The results highlight the limited cross‑cohort robustness and biological consistency of current transcriptomic ICI predictors.
By Yuheng Liang, Lucy Chhuo, Ahmadreza Argha, Nona Farbehi, Lu Chen, Roohallah Alizadehsani, Mehdi Hosseinzadeh, Min Yang, Thantrira Porntaveetusm, Youqiong Ye, Hamid Alinejad-Rokny