Noticias de IA - ítems

arXiv:2605.10981v2 Announce Type: replace-cross Abstract: Reference-free preference optimization has emerged as an efficient alternative to reinforcement learning from human feedback, with Simple Preference Optimization(SimPO) demonstrating strong performance by eliminating the explicit reference model through a simple objective. However, the joint tuning of the hyperparameters $\beta$ and $\gamma$ in SimPO remains a central challenge. We argue that this difficulty arises because the margin formulation in SimPO is not easily interpretable across datasets with different reward gap structures. To better understand this issue, we conduct a comprehensive analysis of SimPO and find that $\beta$, rather than merely serving as a reward scaling factor, primarily functions to filter noisy samples, while the reward margin $\gamma$ defined by SimPO is not semantically well aligned with explicitly enlarging the reward gap between chosen and rejected responses. Motivated by these observations, we propose $\xi$-DPO: Direct preference optimization via ratio reward margin. We first reformulate the preference objective through an equivalent transformation, changing the optimization target from maximizing the likelihood of reward gaps to minimizing the distance between reward gaps and optimal margins. Then, we redefine the reward in a ratio form between the chosen and rejected, which effectively cancels the effect of $\beta$ and yields a bounded and interpretable margin. This margin is called the ratio reward margin and is denoted by $\xi$. Unlike the $\beta$ and $\gamma$ in SimPO, $\xi$ can be determined from the initial reward gap distribution, thus much easier to tune. Our extensive experimental results show that $\xi$-DPO substantially outperforms existing preference optimization methods across open benchmarks on multiple evaluation metrics.

arXiv:2605.10120v2 Announce Type: replace-cross Abstract: Multimodal large language models (MLLMs) show remarkable potential for scientific reasoning, yet their performance in specialized domains such as microscopy remains limited by the scarcity of domain-specific training data and the difficulty of encoding fine-grained expert knowledge into model parameters. To bridge the gap, we introduce MicroWorld, a framework for structured-knowledge augmentation with a complementary visual retrieval pathway, using a multimodal attributed property graph (MAPG) without domain-specific fine-tuning. MicroWorld extracts biomedical entities and relations via scispaCy or LLM-based triplet mining, aligns images and entities in a shared embedding space using Qwen3-VL-Embedding, and assembles a knowledge graph comprising approximately 106K nodes and 2.62M typed edges spanning eight relation categories. At inference time, a graph-augmented retrieval pipeline matches query entities to the MAPG and injects structured knowledge context into the MLLM prompt. On MicroVQA, MicroWorld raises Qwen3-VL-8B-Instruct accuracy from 48.8% to 67.1% (+18.3 percentage points; 37.5% relative), exceeding zero-shot GPT-5 (59.4%). The identical retrieved context raises GPT-5 to 71.5%, and GPT-5's own entity extraction reaches 73.3%. On MicroBench, 8B Single-Pass improves overall accuracy by 4.0 percentage points. The benefits are greatest for knowledge-intensive reasoning; fine-grained visual discrimination and the additional visual-pathway contribution remain limited. Code and accompanying resources: https://github.com/ieellee/MicroWorld.

arXiv:2605.09316v2 Announce Type: replace-cross Abstract: How can a learned representation remain useful when the question it must answer is chosen only after encoding? Preserving information is essential, but does not by itself ensure access to the requested target. We address this problem by deriving "neural information causality (Neural-IC)," a representation law that connects causal access to the physically specified interface capacity. For a fresh database of independent bits, the aggregate information recoverable across later queries cannot exceed that capacity, even with pre-established quantum resources. A decomposition separates capacity slack, information organized jointly across targets, and losses in producing answers. Exhaustive evaluation of trained binary codes reveals why nearly full use of a memory's information budget can coexist with limited retrieval. In a nested Bell-correlation construction, quantum assistance improves average retrieval over the optimal classical one-bit benchmark within the same communicated budget. These results give representation learning a design principle that connects information capacity to the organization and extraction of target information. By locating the constraints on later answers, the framework can inform memory allocation, encoding objectives, and decoder outputs, providing a basis for seeking better retrieval within a fixed budget.

arXiv:2605.07600v2 Announce Type: replace-cross Abstract: Recent methods for improving LLM mathematical reasoning, whether through MCTS-based test-time search or causal graph-guided knowledge injection, cannot identify which concepts causally contribute to a correct answer, as the observed association may be spurious, driven by confounders such as problem difficulty. We propose CIKA (Causal Intervention for Knowledge Activation), a framework that uses the LLM itself as an interventional simulator: a prompt sets the concept state to ``mastered'' and the correctness change estimates the causal effect. We formalize this quantity as an Interventional Capability Probe (ICP), which diagnoses whether the LLM can use a given concept -- distinct from merely possessing knowledge. Because the intervention exogenously sets the concept state independently of problem difficulty, ICP separates confounding that observational methods cannot. On 67 screened problems, the ICP of the top-ranked concept (+0.219) is significantly larger than that of the negative control (+0.039; paired $t$-test, $p < 10^{-6}$, Cohen's $d = 0.86$), confirming that the probe discriminates causally relevant concepts from irrelevant ones. Analysis of 601 Omni-MATH problems further shows that solved problems have 6.1$\times$ higher ATE than unsolved ones (0.338 vs. 0.055), confirming that ICP is predictive of problem-solving success. With a 7B-parameter LLM whose weights are entirely frozen, CIKA achieves 69.7\% on the contamination-free Omni-MATH-Rule benchmark and 64.0\% overall, compared to 60.5\% for o1-mini, and 97.2\% on GSM8K, 46--50\% on AIME 2024--2026, and 46.2\% on MathArena. The Causal Knowledge Activation component contributes 33.8\% of correct answers on problems where the base model alone fails, demonstrating that the LLM already possessed but had not activated the requisite knowledge.

arXiv:2605.06908v2 Announce Type: replace-cross Abstract: Adaptive test-time compute for LLM agents aims to invoke extra computation only when it improves performance. Existing methods typically use confidence-, uncertainty-, or difficulty-based gates, assuming a fixed direction from the gating signal through compute need to the value of computation. This makes gating a utility-calibration problem: gating signals should align with whether extra computation improves the final outcome over the base policy. We show that this alignment is unstable: the same signal predicts rollout benefit in one setting and rollout harm in another, with reversals across environments and backbones even when the task is fixed. Wrong-direction gates can therefore worsen performance by precisely selecting harmful states. This reversal reflects a deeper distinction between compute need and compute suitability: a high uncertainty signal may indicate decision-difficult states where rollouts help compare alternatives, or intervention-unsuitable states where the current context does not support useful rollout-based improvement. Under this two-source model, fixed-direction gates are unreliable across heterogeneous settings. To address this, we propose DIAL (Direction-Informed Adaptive Learning), a sparse gate trained from signal-agnostic counterfactual exploration to learn the utility direction of state features per (environment, backbone). Across six environments and three backbones, DIAL yields a stronger overall success-cost trade-off than fixed-direction baselines. Our code is open-sourced at https://github.com/zmli6/DIAL.

arXiv:2605.06728v2 Announce Type: replace-cross Abstract: Large language models can reason over biological knowledge, but they are not naturally equipped to work with quantitative omics measurements. Existing approaches either convert expression profiles into long textual representations, often losing quantitative information, or connect omics encoders to language models for narrow tasks. We introduce OmicsLM, a 9B multimodal language model that represents each bulk or single-cell transcriptome as a compact continuous embedding that can be interleaved with natural-language instructions and with other samples in the same context. An key contribution is a training corpus that connects experimental measurements with biological knowledge and task supervision. We combine templates linking transcriptomic profiles to annotations and experimental outcomes, reasoning examples connecting evidence across sources, and LLM-generated supervision with task-specific quality checks. The instruction-tuning corpus spans 157 biological and general-purpose task groups. Training includes continued pretraining on 20 billion tokens and supervised fine-tuning on 5.21 million conversations. We further introduce GEO-OmicsQA, a publication-disjoint benchmark of 3,000 questions requiring models to interpret and compare expression profiles from real GEO studies. OmicsLM achieves the best results among the evaluated models on three of four held-out annotation tasks and on all six GEO-OmicsQA variants, and the best average performance among compared methods on PerturbQA perturbation-response prediction. Ablations show a clear advantage for continuous omics representations over a text-only gene-list representation. Together, these results show that quantitative transcriptomic measurements and biological language can be integrated within a general conversational model, enabling both predictive tasks and language-guided interpretation of experimental samples.

arXiv:2605.05164v2 Announce Type: replace-cross Abstract: Accurate analysis of histopathological images is critical for disease diagnosis and treatment planning. Whole-slide images (WSIs), which digitize tissue specimens at gigapixel resolution, are fundamental to this process but require aggregating thousands of patches for slide-level predictions. Multiple Instance Learning (MIL) tackles this challenge with a two-stage paradigm, decoupling tile-level embedding and slide-level prediction. However, most existing methods implicitly embed patch representations in homogeneous Euclidean spaces, overlooking the hierarchical organization and regional heterogeneity of pathological tissues. This limits current models' ability to capture global tissue architecture and fine-grained cellular morphology. To address this limitation, we introduce a hybrid hyperbolic-Euclidean representation that embeds WSI features in dual geometric spaces, enabling complementary modeling of hierarchical tissue structures and local morphological details. Building on this formulation, we develop BatMIL, an MIL method that leverages both geometric spaces. To model long-range dependencies among thousands of patches, we employ a structured state space sequence model backbone that encodes patch sequences with linear computational complexity. Furthermore, to account for regional heterogeneity, we introduce a mixture-of-experts module that dynamically routes each patch to specialized expert sub-networks based on its features, improving representational capacity while reducing redundant computation. Experiments on seven datasets spanning six cancer types demonstrate that BatMIL achieves outstanding performance against fifteen state-of-the-art MIL approaches, demonstrating that geometry-embedding representation learning offers a promising direction for computational pathology.

arXiv:2605.04971v2 Announce Type: replace-cross Abstract: Weight matrices in deep networks exhibit geometric continuity---principal singular vectors of adjacent layers point in similar directions. While this property has been widely observed, its origin remains unexplained. Through experiments on toy MLPs and small transformers, we identify two mechanisms: residual connections create cross-layer gradient coherence that aligns weight updates across layers, and symmetry-breaking nonlinearities constrain all layers to a shared coordinate frame, preventing the rotation drift that would otherwise destabilize weight structure. Crucially, a nonlinear but rotation-equivariant activation reaches less than half of ReLU's continuity, isolating symmetry breaking---not nonlinearity alone---as the active ingredient. Activation and normalization play distinct roles: activation concentrates continuity in the leading singular direction, while normalization distributes it across multiple directions. In transformers, continuity is projection-specific: Q, K, Gate, and Up (which read from the residual stream) develop input-space ($\mathbf{v}_1$) continuity; O and Down (which write to it) develop output-space ($\mathbf{u}_1$) continuity; V alone, lacking an adjacent nonlinearity, develops only low continuity.

arXiv:2605.02525v2 Announce Type: replace-cross Abstract: Semantic indoor navigation requires robots to connect human instructions with spatial destinations, execute missions and retain useful operational experience. This paper presents the Semantic Autonomy Stack (SAS), a six-layer integration framework, and develops its semantic-reasoning, mission-interface and operational-memory layers over a previously validated ROS 2/Nav2 foundation. A deterministic-first cascade combines the route graph, semantic objects, robot state and stored instruction-destination associations, escalating unresolved or guarded cases to a pretrained Vision-Language Model (VLM). Recurrent model-mediated associations can pass a frequency-and-consistency gate and be compiled for deterministic M3 reuse; this gate measures sampled-output stability rather than semantic truth. The frozen v4.8-review implementation was evaluated on two physical robot configurations in one indoor environment, with onboard navigation and perception and shared-workstation reasoning and memory. Of 216 decision cycles outside the start gate, 112 resolved deterministically at sub-millisecond resolver-compute latency. A sealed 28-instruction challenge yielded 17 acceptable outcomes; five fast-path decisions included one incorrect selection. One promoted association was reused on the second platform through deliberate shared-digest installation. A controlled 594-observation offline comparison showed greater agreement with the stored associations and lower decision latency from M3 reuse in five familiar cases. The M3-disabled cascade reduced VLM calls but matched direct-VLM decision quality in standardized context. Large latency savings were concentrated in paths avoiding VLM inference. Together, the results establish a working integration of selective hybrid reasoning and reusable operational memory, while identifying implementation and evaluation boundaries for further development within SAS.

arXiv:2605.01720v4 Announce Type: replace-cross Abstract: Large-scale sign language resources usually pair RGB videos with text and support recognition, translation, and generation. However, RGB frames contain backgrounds, clothing, and other appearance information unrelated to signing motion. Representing the videos as whole-body keypoints such as DWPose removes much of this variation, produces a compact description of body, hand, and facial movement, and matches the input used by recent pose-guided video models. Existing datasets do not provide this representation at scale for in-the-wild sign language videos. We present SignVerse-2M, a multilingual dataset containing approximately two million pose clips from more than 55 sign languages. We process publicly available sign language videos with a shared DWPose pipeline, representing body, hand, and facial motion in the same keypoint format across all clips. The source videos include diverse signers, viewpoints, and recording environments. We also release the processing pipeline, task definitions, and SignDW Transformer baselines. Experiments show that the data supports multilingual text-to-pose modeling and that the generated poses can directly drive several recent pose-conditioned rendering systems.