Papers

In-context watermarking (ICW) prepends an instruction to a query asking the model to embed a statistically detectable signal in its response. It thus equips LLMs with a watermarking interface that third parties can invoke without access to model internals. Its reliability hinges on the LLM following the instruction without degrading answer quality, yet how well current LLMs do so has not been measured. We introduce $\mathsf{ICWBench}$, a benchmark of three verifiable ICW instruction families, each scored on both detectability and answer quality. Evaluating 14 frontier proprietary and open-source LLMs, we find that none of the evaluated LLMs achieves both objectives across all three families. To address this, we propose a self-contained two-stage training method, requiring no distillation from a stronger model, no manual annotation, and no pre-existing ICW IF ability. The first stage, self-distillation with logits perturbation (SDLP), uses the same base LLM as both teacher and student: an instruction-equivalent decoding-time logits perturbation makes the teacher follow the ICW instruction, and the student is trained to match the teacher's output distribution. The second stage applies reinforcement learning with the automatic verifier as the reward. Applied to Qwen3-14B and GPT-OSS-20B, our method raises average TPR@$1\%$FPR across three ICW instructions from $0.100$ to $0.974$ and from $0.337$ to $0.968$, respectively, while maintaining high response quality under both perplexity evaluation and LLM-as-a-Judge.

Joint-Embedding Predictive Architectures (JEPAs) provide a powerful framework for latent world modeling and planning in a reconstruction-free manner. Although numerous JEPA-based approaches have been proposed to mitigate representation collapse, our experiments on localized, out-of-distribution visual noise reveal that performance degradation remains pronounced and unresolved. We propose Flow-JEPA (F-JEPA), a flow-based latent dynamics model that jointly generates a sequence of future latent states conditioned on the current observation and actions. A Gaussian distribution serves as the flow source, exposing the vector field to perturbed latent trajectories as it learns to transport them toward clean future representations. This formulation retains the reconstruction-free JEPA framework while switching from pointwise transition regression to stochastic trajectory-level prediction. F-JEPA raises mean success from $86\%$ to $92\%$ under clean observations and from $67\%$ to $86\%$ under noisy conditions. Further evaluations over varying perturbation severity and inference settings show that the performance advantage persists across a broad range of conditions. These results suggest that conditional flow matching provides a promising alternative to deterministic autoregressive prediction as a dynamics formulation in JEPA world models.

Multi-agent LLM systems often coordinate by compressing an upstream interaction into a handoff artifact that downstream agents treat as shared state. We show that this handoff step is a structural source of privacy leakage: summaries preferentially preserve operational facts while weakening the boundary metadata that governs how those facts may be used—a failure mode we call summary collapse. On a controlled multi-agent coordination testbed we measure marker survival with a human-validated judge ($\kappa = 0.74$), where $\sigma_b = 1$ means every boundary marker survives verbatim and $\sigma_b = 0$ means all are lost. Boundary-marker and operational-fact survival are nearly uncorrelated at the handoff level on both GPT-5-mini and DeepSeek-R1-32B (Pearson $r$ near zero): uncompressed free-text handoffs preserve boundaries at $\sigma_b \approx 0.80$, whereas a $25$-word budget drops $\sigma_b$ to ${\approx}0.57$ while operational-fact survival stays near ceiling. Controlled downstream tests reveal that protection depends on boundary explicitness: vague languages leak in $73\%$ of GPT and $50\%$ of DeepSeek cases, while explicit constraints reduce leakage to under $15\%$ across all three tested models. A no-handoff single-agent control further shows the failure is not reducible to multi-agent topology as direct full-marker access still leaks more often than the operationalized handoff. Prompt-only mitigation and exact-string redaction only partially address the problem, while a gold-derived audience allowlist nearly eliminates leakage across models, showing that correctly identifying audience boundaries is the key factor.

Deep learning has a powerful capability of feature extraction. However, the lack of fairness and interpretability in deep neural networks poses limitations to their adoption in the medical domain. This paper proposes a disentangled representation learning (DisenRL) framework, named the Attributes-based Gaussian Estimation for Disentangled Representation (AGEDR), which incorporates Attribute Mapping Embedding (AME) modules designed to map attributes into vectors and align them with a subset of the latent vectors in a Variational AutoEncoder (VAE). This part of the latent vector will be disentangled from the remaining latent vectors by minimizing mutual information. A classifier is then trained using the mean parameters of the latent vectors from the VAE. Extensive experiments demonstrate that AGEDR outperforms both conventional classification models and existing disentangled representation learning methods. The ablation experiments also indicate the disentangling capability and fairness of AGEDR. The source code is publicly available at https://github.com/ZhaoKe1024/DisentangledRepr.

Time-series data across physical and biological domains are fundamentally driven by complex, non-stationary oscillatory modes. While deep learning models, such as Convolutional Neural Networks (CNNs), Recurrent Neural Networks, and Transformers, have dominated sequential analysis, they remain fundamentally "spectral-blind". By mapping continuous physical waves into unconstrained spatial or discrete token spaces, these architectures suffer from severe spectral entanglement, acting as opaque black boxes that decouple predictive accuracy from physical reality. In this paper, we introduce the Frequency Selective Neural Network (FSNN), pioneering a foundation architecture guaranteeing physical interpretability without sacrificing expressive power of deep learning. FSNN addresses spectral entanglement by explicitly embedding the rigorous mathematics of advanced signal processing into its neural topology. Through a fully differentiable Wiener-like filter bank optimized via complex-domain backpropagation, FSNN autonomously discovers and isolates the precise physical modes of a given task. Extensive evaluations demonstrate that FSNN establishes state-of-the-art predictive performance, achieving $77.0\%$ average accuracy on the standard 10 multivariate UEA datasets and leading across all major metrics on the highly imbalanced PTB-XL clinical ECG benchmark. Crucially, in contrast to yielding abstract feature maps, FSNN converges directly on physically meaningful frequency bands, such as isolating the cardiac QRS complex, providing a highly scalable, interpretable paradigm for robust pattern recognition in complex temporal domains. Our code is available at: https://github.com/ad6174hhhh/FSNN.

Accurate forecasting of grape berry temperature (Tb) is essential for enabling timely heat stress management in vineyards. In this study, a feed-forward attention mechanism integrated with a Long Short-Term Memory network (FAM-LSTM) was developed and evaluated for multi-step, high-resolution Tb prediction. Models were trained using environmental data from 2023 and 2024 at Prosser, WA, USA, and validated on 2025 summer data. FAM-LSTM was benchmarked against LSTM, GRU, RNN, and Random Forest (RF) across horizons ranging from 15 minutes to 72 hours (288 time steps). Two input scenarios were evaluated: nearest open-field weather station observations and in-vineyard microclimate measurements. FAM-LSTM consistently outperformed all benchmark models across all horizons and input scenarios. Incorporating in-vineyard microclimate data significantly improved forecasting accuracy at longer horizons. Using open-field data, FAM-LSTM achieved MAE and RMSE ranges of 0.58 to 1.70 deg C and 0.65 to 2.07 deg C, respectively. In-vineyard observations further improved performance, with MAE and RMSE in the ranges of 0.51 to 1.55 deg C and 0.71 to 1.87 deg C. Error analysis showed prediction uncertainty was highest during peak daytime periods (11:00 to 18:00) and increased progressively with forecast horizon. Overall, the FAM-LSTM framework offers robust Tb forecasting to support precision heat stress management in vineyards.

In dynamic and unstructured environments, conventional SLAM systems generally suffer from significant accuracy degeneration due to their static assumptions. In this work, we propose Robust Semantic-aware Gaussian Splatting SLAM (RoSe-SLAM), to address the dynamic challenge by a holistic semantic scene understanding from uncalibrated monocular inputs, achieving accurate camera tracking and high-quality geometry reconstruction. Unlike conventional semantic SLAM using handcrafted semantic labels, our RoSe-SLAM exploits the semantic feature from 2D foundation model to enhance the dynamic tracking and mapping performance. By distilling the rich semantic features to our Gaussian fields, our method effectively identifies dynamic distractors and achieves semantic-aware multi-view consistency, significantly enhancing the geometric reconstruction and scene inpainting. Specifically, we propose a spatial-temporal motion mask generation module, enabling both long-term motion monitoring and short-term transient dynamics capturing, achieving robust and effective disentanglement of dynamic objects and static backgrounds. During global bundle adjustment, we propose an occlusion-aware keyframe selection mechanism to prioritize the occlusion as metric to pick the keyframes, and a multi-view semantic consistency module to improve the mapping quality in dynamic environments. By combining geometric motion cues with semantic priors, our system dynamically filters unreliable observations and reconstructs accurate static scene geometry. Extensive experiments conducted on benchmark datasets including dynamic TUM, Bonn and Wild-Mocap datasets, demonstrate that our method achieves superior performance in both trajectory estimation and static scene mapping, outperforming existing dynamic RGB SLAM baselines in long-term dynamic indoor environments.

In May 2026 an OpenAI model produced a counterexample to the Erdős unit distance conjecture. Five mathematicians published a human-verified version the same day, and the result entered the literature within weeks. In August 2026 the same laboratory published ten mathematical and theoretical computer science results, each accompanied by a machine-checkable Lean 4 certificate with no unproved steps. Four weeks later, one remained the subject of an unresolved dispute over whether its formalization meant what it claimed. We argue that this difference is structural. We distinguish three layers of verification: derivational validity, which a kernel checks; representational fidelity, whether the formal statement means the intended question; and epistemic significance. Only the first is mechanizable. Making it effectively free therefore does not eliminate verification work but shifts the burden to layers dependent on scarce expert attention. Measurements of the August corpus illustrate the shift. The kernel-checked proofs total 20.6 MB, while the statements requiring human audit total 55.6 KB, a ratio of 379 to 1. Yet those statements contain 218 bespoke definitions rather than relying on community-vetted ones. The audit surface is therefore small in volume but irreducibly expert. We argue that machine checking produces verification abundance while leaving adjudication scarce. We propose a six-category taxonomy of representational mismatch, a disclosure schema for machine-generated mathematical claims, and implications for software, cryptography, and regulated decision systems.

Protein co-abundance clusters preserved across tissues can reveal shared disease mechanisms and candidate therapeutic targets, particularly when proteins implicated in organ-confined diseases converge in peripheral or accessible tissues. However, previous cross-tissue studies have focused on biologically pre-selected tissue pairs, leaving most possible combinations and non-obvious relationships unexplored. We present an LLM-agent framework for large-scale, evidence-grounded comparison of tissue-specific protein co-abundance networks. The framework constructs tissue networks, derives pairwise consensus clusters, and integrates evidence from expression atlases, protein interaction and complex databases, pathway annotations, disease catalogues, and literature. Applied to all 820 pairwise combinations of 41 human tissues and fluids, it identified 1,833 conserved co-abundance clusters across 406 tissue pairs. Colon, synovial fluid, blood, cerebrospinal fluid, and bone marrow were the most broadly connected tissues, while the most cluster-rich pairs were dominated by bone marrow. The analysis also highlighted non-obvious relationships: skin-bone marrow exceeded the anatomically adjacent bone-bone marrow pair, while colon-breast contained cancer-relevant clusters involving extracellular-matrix remodeling, lipid metabolism, and immune modulation. Cluster-level analyses generated further mechanistic hypotheses, including a brain-gut extracellular-vesicle/redox/serotonin-cofactor axis and a liver-bone marrow stress-response axis involving genes linked to white matter disease. These results provide a global, comparable landscape of conserved protein co-abundance and a hypothesis-generating resource for mechanistic and therapeutic exploration. Code and data are available at https://github.com/Gry1005/AgenticAI-conserved-cross-tissue-protein-co-abundance.

Specialized machine learning architectures encode structural assumptions (equivariance, permutation invariance, relational structure) that language-based foundation models lack by design. This review asks whether such assumptions can instead be acquired through language, drawing on a corpus of 186 papers published between 2016 and 2026 across nine modalities: tabular data, graphs, time series, vision, chemistry, code, knowledge graphs, point clouds, and protein structure. Methods are organized into eight representational regimes, ranging from language-only prompting to fully specialized architectures, and are assessed not only on predictive accuracy but on whether structural information is preserved by the representation and computed by the model. Language-mediated models prove competitive in extreme few-shot prediction, discretized symbolic tasks, and textually annotated knowledge graphs. No evidence of general replacement survives once structure itself is evaluated: where direct tests exist, apparent parity is traceable to information asymmetry, benchmark contamination, or computation performed outside the language model. Across independent research communities, missing structural inductive biases are reintroduced through graph modules, structure-aware tokens, or specialized attention, indicating that specialization is relocated rather than eliminated. The review closes by specifying a falsifiable controlled experiment that would measure the remaining gap directly. The derived data supporting the findings of this review are openly available in the repository at https://github.com/kiyan-rezaee/language-vs-structure.

Knowledge graphs have been proposed as a structured alternative to flat retrieval-augmented generation for long-term agent memory, on the assumption that representing conversations as entities and relations improves recall. We evaluate that assumption directly. Our framework extracts each conversational turn into typed nodes and attributed edges, answers questions from a two-hop subgraph, and periodically prunes nodes that score low on a weighted combination of recency, access frequency, degree centrality, and age. On LongMemEval, the graph does not outperform a flat vector baseline at a matched candidate-generation budget of five retrieval roots: token F1 is $0.417$ against $0.468$, and a paired bootstrap over 500 questions gives $\Delta = -0.050$ (95% CI $[-0.085, -0.016]$). The gap is widest on questions that require recalling a specific prior assistant turn, where judged correctness falls from $0.911$ to $0.607$, suggesting that decomposing a turn into entities discards the surface form these questions depend on. The forgetting module is more successful. Applied once to a persistent 27{,}021-node graph, it removes 9.8% of nodes and 9.5% of stored bytes; token F1 is unchanged ($+0.001$, 95% CI $[-0.015, +0.016]$) and judged correctness falls by $1.6$ points, with the 95% interval bounding any loss at $3.8$ points ($[-0.038, +0.006]$). Because our extractor is a single small model evaluated on one benchmark, these results characterise this extraction-based pipeline rather than graph-structured memory in general. Code: https://github.com/skhanzad/Selective-Amnesia

Several works have investigated the influence of graph topology on cooperation among artificial agents, while the majority of the literature has focused on modelling agents' adaptation through strategy imitation, which relies solely on the cumulative payoffs of others. This paper investigates scenarios in which each agent learns to play the two-player Iterated Prisoner's Dilemma (IPD) using deep reinforcement learning. Each agent is represented as a node in a graph, where its neighbours constitute the pool of opponents with whom it can interact. During each IPD episode, agents are provided with different types of information about their opponent, consisting of action history and opponent identity. Experimental results across different graph topologies show that the number of neighbours per node and the average path length are the main factors affecting the emergence of cooperation. We also show that, while partner selection fosters mutual cooperation by limiting the diversity of the opponent pool, providing agents with the identity of their opponent hinders the proliferation of cooperative strategies.

Clinically relevant oncology information is distributed across heterogeneous, longitudinal documentation, creating substantial abstraction burden and requiring accurate attribution across specimens, tumors, biomarkers, and time points, while manual cancer-registry abstraction can require 27.2 minutes per case, highlighting the need for scalable methods that preserve clinical context while converting documentation into structured data. We evaluate an oncology information-extraction workflow in which OncoLens supplies multi-source, oncology-aware document selection, aggregation, and normalization from integrated EHRs, while the NimbleMind Multi-Agent System (nMAS) is a configurable oncology information-extraction workflow that extracts clinically relevant structured fields from fragmented oncology documentation. The extraction task uses a clinician-informed schema of 328 attributes spanning report metadata, diagnosis, staging, and cancer-type-specific information. nMAS separates clinician-defined field specifications from model execution and combines complexity-aware extraction, report-level consolidation, and source-grounded validation. The retrospective evaluation included 230 de-identified oncology documents from 40 patients and 418 clinician-reviewed document-field pairs containing 1,126 non-empty reference values. Evaluation focused on fields identified by clinicians as present in the source documents rather than exhaustively annotating all 328 schema fields. nMAS achieved a rank-weighted value-level precision of 82.6%, recall of 87.5%, and F1 of 85.0%, compared with an F1 of 66.4% for an independently implemented UMA-style MiniMax M2.5 comparator. These findings support the feasibility of using a configurable, source-grounded extraction workflow to convert fragmented oncology documentation into reusable structured data.

Philosophers and legal scholars are engaged in debates about the implications of artificial intelligence for freedom of expression. This paper analyzes the free speech issues raised by two distinct AI technologies: social media recommendation algorithms and conversational AI (i.e., chatbots powered by large language models). The first part shows that, through their recommendation algorithms, social media platforms control the dynamics of speech visibility in the digital public sphere, making algorithmic recommendation relevant to the philosophy of free speech. The second part turns to conversational AI. It discusses both the reasons for granting or withholding speech rights to artificial agents and users' right to receive information, which may render specific forms of chatbot regulation illegitimate. Throughout, the chapter also considers whether social media platforms or AI developers hold corporate speech rights. Its general aim is to raise rather than settle questions that arise from the rapid development of AI technologies.

Current TTS systems typically rely on open-loop, single-pass generation and can produce sporadic local prosodic defects, such as misplaced stress, unnatural pauses, or flattened intonation, that utterance-level metrics often fail to expose. We present LoopTTS, a judge-guided Filter-Judge-Refiner framework for recovering low-quality TTS outputs diagnosed by an AudioLLM. Given an initial utterance from a base TTS model, an AudioLLM Judge identifies salient prosodic issues and generates structured refine instructions; a Refiner, our fine-grained instruction-following TTS model, then performs guided expressive re-synthesis conditioned on the initial utterance, target text, and instruction. To train the Refiner, we construct Refiner-DB, a 42K-example AudioLLM-annotated dataset with word-level prosodic weak supervision. Human evaluation on diagnosed low-quality utterances shows that LoopTTS can detect perceptually salient errors and correct them with the Refiner, outperforming raw generated audio and practical open-loop re-generation baselines in recovery quality. The Refiner also demonstrates stronger instruction-following ability for stress and pause control in targeted prosody modification.

Semantic Search on LinkedIn must retrieve relevant profiles from a corpus of hundreds of millions in response to natural-language queries such as "a fintech founder in Berlin who worked in payments." The deployed relevance policy is bottleneck-oriented: every active non-negotiable facet must be satisfied, and a pre-existing LLM Graded Relevance (GR) judge operationalizes this through a fixed min/median aggregation over facet grades. Cosine similarity instead averages evidence, letting a strong match on one facet mask failure on another, capping the recall of the first-stage (L0) retriever. We present a policy-aligned retrieval framework: embeddings are partitioned into eight category-supervised segments whose scores follow the same min/median rule at serving time; for multi-vector retrieval, this segment score is computed independently per tagged document slot and maximized across slots. A lightweight single-slot Stage-1 scorer generates high-recall candidates, while scale-invariant relative-norm gating keeps category activation consistent across training, evaluation, and serving. On 21K held-out queries, this representation improves offline relevance over a matched-capacity baseline, with gains broadly distributed across facet combinations. We serve this framework with a two-stage GPU architecture: an FP8 coarse ranker scores the full corpus, increasing per-shard capacity by 71% and Stage-1 matmul throughput by 36%, then an FP16 stage exactly re-ranks an oversampled candidate set, recovering 99.6-99.8% of full-FP16 recall at over 500 QPS per shard replica. In a member-randomized A/B test, exploratory-query Precision@10 under the unchanged GR judge rises from 63.7% to 79.0% and navigational Precision@1 from 65.5% to 74.7%, with a blinded human evaluation independently confirming the Precision@10 gain.

People search must map free-form location phrases to geographic entities used as structured retrieval filters. Lexical standardizers handle canonical names well but are brittle to aliases, misspellings, metropolitan expressions, and same-name ambiguity. We formulate this task as graded, set-valued entity retrieval over a fixed ontology. We identify three coupled design requirements: distinguishing identity-preserving variation from knowledge-dependent aliases, controlling false negatives among valid same-name entities, and separating stable transformations from mutable entity knowledge. We realize them in a prompt-asymmetric bi-encoder with calibrated alias support, bounded ambiguity-aware negatives, and editable entity documents that support localized updates without retraining. Across a fixed production-derived development benchmark and a public GeoNames transfer task, task adaptation improves substantially over frozen encoders and standard token baselines. Controlled development ablations show that specialized supervision contributes beyond standard task fine-tuning and encoder scaling. On GeoNames, the adapted model improves known-target Recall@1 throughout zero-to-moderate character overlap, while character n-grams retain a small aggregate Target Recall@5 advantage. In a blinded human comparison on a stratified production challenge set, our model raises relevant P@1 from 28.0% to 46.0% (p=0.012). Fixed-query endpoint estimates improve on non-canonical queries and remain close to control on frequent queries; a randomized live experiment detects no engagement regression. These results support task-adapted geographic entity retrieval as a practical replacement for the incumbent taxonomy-based standardizer, with the largest relevance gains on non-canonical queries.

Verified search, in which a language model proposes programs, a hard evaluator scores them, and selection keeps the best, has recently moved mathematical records; controlled ablations of the proposer-side components remain rare. We instrument a minimal FunSearch-style loop at office scale (a 30B local model on a laptop, 120-600 verified samples per run) with three operator packages: a schematic notebook the model writes and carries instead of verbatim elites, a named obstacle, and behavioural repulsion from constructions already found. On nine construction problems from a public repository, the complete 2^3 factorial with two replicates favours the primary contrast in a nominal two-stage analysis: the composition closes more of the seed-to-record gap (+0.196; nominal pooled p=0.023, stage-combination p~0.08; median per-problem effect +0.045). Repulsion raises construction-hash diversity in eight of nine problems (one-sided p=0.0039; partly a manipulation check). The factorial finds no positive memory-by-repulsion interaction (bounded to about +/-0.04); the gain decomposes additively, and memory+repulsion is the only arm that never collapses (0 of 18 runs), within 0.025 of the full composition. A frontier proposer under the identical loop reaches in tens of samples what the local model does not in hundreds; in single scoping runs its gains arrive without the operators. The searches stall after closing 85-92% of the gap on the flagship problem, and a later analysis reads that plateau as a resolution limit: the reference value is a continuum limit and the unaided long-run finals already carry its family's structure. In the registered family-hint test the family named in words is adopted and loses; handed as code, it is optimized, but our best finite-grid encoding stays below the plateau reached unaided. We release the harness, every candidate, and the dated pre-registrations.

Recursive AI, the prospect of AI taking an increasingly complete role in building and improving AI, is a crown jewel of AI for AI. Although recursive self-development has become practical for small models, bounded tasks, and fixed time budgets, a more consequential realization of this ambition, i.e., developing a release-ready, frontier-competitive model, remains far more challenging. In this work, we ask how little human involvement is sufficient for an agent to develop a frontier model. We concentrate human input into a high-density, low-frequency interface: experts encode objectives, stage scaffolds, permission boundaries, and operating procedures as reusable research skills, while the agent instantiates these priors, selects experiments, diagnoses outcomes, and revises the training strategy. In the challenging domain of industrial coding, the agent evolves data and coordinates SFT, on-policy self-distillation, and reinforcement learning with verifiable rewards, ultimately producing iCoder, a 27B model for RTL design and GPU kernel optimization. Across seven benchmarks, iCoder leads RTLLM, outperforming GPT-5.5 and Claude-Opus-4.8; ranks second on CVDP and KernelBench L2, exceeding GPT-5.5 by 16 points; and ties Claude-Opus-4.8 for the best TritonBench result. Exploratory case studies further show iCoder's competitive iterative RTL and GPU-kernel optimization with substantially fewer tokens. These results chart an engineering path toward recursive self-improvement, in which humans distill the principles of model building, agents operationalize them through evidence-driven experimentation, and each generation of AI becomes a more capable architect of the next.

We report an exploratory, single-seed ablation of TALH (Adaptive Latent Hybrid), a decoder-only language model with parallel Multi-head Latent Attention (MLA) and a custom recurrent state-space (SSM) branch. Five variants, spanning 117–217M estimated active parameters per token, are trained from scratch on a FineWeb sample for the same number of optimisation steps and tokens. In this specific setup, removing the SSM branch gives the largest degradation in validation perplexity (MLA-only PPL 315), whereas removing MLA has a much smaller effect (SSM-only PPL 239). A dense-FFN hybrid obtains PPL 231, compared with 240 for the tested top-2 ternary-MoE hybrid, while using 3.87 GB less peak training memory. We also preserve a preliminary Apple M3 timing observation: among the five unoptimised implementations, MLA-only has the flattest measured time-to-first-token curve from 512 to 2,048 prompt tokens, although the dense Transformer is much faster in absolute terms. Because the runs are single-seed, parameter counts are unmatched, the evaluation stream may overlap the training source, and raw repeated timing records are unavailable, these results support implementation-specific hypotheses rather than general conclusions about MLA, SSMs, or mixture-of-experts models.

We present a modular and multi-level framework for the differential diagnosis of malignant melanoma. Our framework integrates contextual information and evidence at the lesion, patient, and population levels, enabling decision-making at each level. We introduce an anatomic-site aware masked transformer, which effectively models the patient context by considering all lesions in a patient, which can be variable in count, and their site of incidence. Additionally, we incorporate patient metadata via learnable demographics embeddings to capture population statistics. Through extensive experiments, we explore the influence of specific information on the decision-making process and examine the tradeoff in metrics when considering different types of information. Validation results using the SIIM-ISIC 2020 dataset indicate including the lesion context with location and metadata improves specificity by 17.15% and 7.14%, respectively, while enhancing balanced accuracy. The code is available at https://github.com/narenakash/meldd.

Agentic AI is sold on a simple promise: the system keeps acting when the user stops watching. That promise creates an accountability inversion. As stepwise supervision recedes, verification does not disappear; it moves into the runtime infrastructure that defines authority, records action, interrupts execution, checks outcomes, and supports repair. We call this the reduced-supervision paradox. Using a 63-artifact audit, we examine its public visibility across 46 research papers and 17 engineering, documentation, security, and governance sources. We find that agents' action surfaces are far easier to reconstruct than the mechanisms needed to answer for their actions. Tool mediation and monitoring traces were clearly visible in 40 and 37 artifacts, whereas checkpoint placement was clearly visible in 6, validator independence in 4, recovery in 2, and contestability in 1. Three action paths show why this imbalance matters. A repository path can preserve rich diffs after a consequential change. A browser path can cross organizational boundaries faster than permissions travel. A service path can follow policy while leaving affected people without recourse. We argue that observability can become a substitute for accountability when it shifts verification onto users after meaningful intervention is no longer possible. Our action-path diagnostic instead asks whether a delegated action remains connected to authority, evidence, interruption, independent judgment, recovery, and challenge. The claim is deliberately bounded to public visibility; it does not establish the prevalence or effectiveness of undisclosed controls. We contribute an action-level account that recasts the harness from a technical wrapper into accountability infrastructure.

Drug-target relation prediction supports candidate screening, drug repositioning, and mechanism analysis. Existing models often use incomplete drug or protein representations, model cross-modal interactions shallowly, or train affinity regression and interaction classification separately, although these tasks describe closely related views of the same drug-target pair. This paper presents DPTM-DT, a dual-pretrained Transformer framework for multitask drug-target prediction. DPTM-DT combines GROVER molecular graph embeddings, ESM protein language-model embeddings, and CTD physicochemical descriptors, then exchanges drug-target information through bidirectional cross-modal attention. A shared pair representation is used for continuous affinity regression, high-affinity binary classification, and six-level affinity classification. Experiments on Davis and KIBA cover random 80/20 and DeepDTA-style standard splits. On the random 80/20 split, DPTM-DT achieves MSE/CI values of 0.193/0.917 on Davis and 0.120/0.918 on KIBA. It also reports binary AUPR/MCC values of 0.727/0.654 and 0.798/0.689, and six-class Macro-F1/Top-2 values of 0.800/0.932 and 0.815/0.962 on Davis and KIBA, respectively. Across the reported regression, binary classification, and multiclass classification settings, DPTM-DT achieves the best overall performance among the compared methods. Results under the standard split show the same relative trend. Ablations indicate that dual target representation, gated fusion, and cross-modal attention each contribute to the final performance. Code and supplementary materials are available at: anonymous.4open.science/r/DPCM-DT-74E0.

Mathematical communities work with different objects, invariants, and tools, so transferring a problem across them is expensive and often skipped. We present EULER, a multi-agent system that takes such a transfer–a bridge–as its unit of search. Around a fixed conjecture, EULER runs direct, adjacent-domain, and distant-domain routes in competition; a bridge keeps its budget only if it supplies an operation the source representation cannot execute and its target-side evidence returns to the original statement along a checked implication. Six ordered stress tests reject invalid bridges before expensive search begins. We evaluate EULER on 120 recent conjectures. The conjectures were frozen before search and screened for contamination, and are drawn from public papers by authors who had recently published in the Journal of Combinatorial Theory, Series A, a leading journal in combinatorics. EULER produced 10 proofs and 3 refutations, plus 45 scoped partial results. Two mechanisms held up under ablation: bridge-specific stress tests cut incorrect conclusions from 9 to 3, and bridge material combined with a target-native operation yielded a positive interaction of +4.2 resolved tasks that neither factor produced alone. Domain distance did not reliably predict success; executable operation gain and valid return did.

We study a class of product-reference diffusion algorithms for sampling from a discrete distribution. We show that their sampling performance can be characterized using a path-based measure of data geometry that we call the interaction growth complexity (IGC). We show that a bivariate IGC kernel gives an exact representation of both the KL discretization error and a simple one-step upper bound. The simpler univariate IGC density can be used to study the effect of stepsize choices on the iteration complexity required to obtain $\epsilon$-accurate samples in KL divergence. Samplers that traverse the path with equi-spaced steps in log-squared-reliability-odds have performance that depends on the aggregate IGC mass, whereas refined choices of stepsizes have a lower complexity depending on a square-root functional. In the fine-grid limit, both of these characterizations become sharp. We also allow general product reference distributions and show that the reference law can substantially reshape the IGC profile and the resulting sampling complexity; in particular, references far from both the uniform and the data marginals can yield dimension-dependent improvements. Finally, the aggregate IGC mass admits bounds in terms of total correlation and dual total correlation, thereby connecting the pathwise geometry to classical measures of multivariate dependence.

Over the course of the last decade, neural networks have grown from an academic curiosity to moving the markets of nations. Despite this explosion in both research and deployment, relatively little is understood about how they achieve the solutions they do. This is both scientifically relevant, and pressing for society. When neural networks make decisions across self-driving, construction, law, hiring and health, there have been and will continue to be unintended consequences. However, attempting to generalize the failures of the largest and most important production systems makes for a very difficult task. Yet signs of these failures exist at all scales of neural networks, so we should be able to study a much more tractable setting. All neural networks must undergo an optimization process, called training, to be useful. To a great degree, understanding neural networks is understanding their optimization: through what process and exposure to which data did they arrive at their results. Yet our knowledge on this topic as a field is quite imprecise. In particular, a curious phenomenon called mode connectivity, the ability to connect neural networks in the loss surface, defies explanation entirely. This dissertation elucidates, explains and exploits this special structure in the loss landscape...

Automating alignment research may accelerate progress toward aligned AI, but whether it does is hard to measure. Luckily, many alignment failures, such as deception, sycophancy, and jailbreaks, are already measurable by public benchmarks. We study whether automated alignment researchers (AARs) can post-train to mitigate alignment failures by proposing training methods and data to simultaneously optimize multiple safety benchmarks, while largely preserving general capability. Across 10 alignment failures, the strongest AAR methods significantly reduce the targeted alignment failures and generalize to a held-out benchmark, multi-turn behavioral audits, and models up to 4.7x larger than the target model. As a human baseline, 28 experienced researchers receive up to eight hours to develop one-shot methods for the same benchmarks, but their methods underperform the best AAR methods. Using human ideas as the AARs' initial research direction does not improve performance, suggesting current AARs may not need guidance from experienced researchers. These results suggest that automating alignment research on well-characterized failures may be practical in the near term.

Credit risk analysis in financial institutions traditionally requires analysts to manually write SQL queries, run statistical computations, and build visualization dashboards. This is a time-consuming workflow that limits exploration to familiar segments. We introduce Oculi, a conversational platform that transforms natural language questions into comprehensive credit risk analyses, complete with data queries, statistical testing, and interactive visualizations. Oculi employs a three-layer architecture that separates reasoning (LLM-powered agent), execution (Model Context Protocol tool servers), and presentation (agentic UI), enabling analysts to discover high-risk portfolio segments. Within Oculi, a new segment discovery pipeline is proposed that combines deterministic statistical methods with LLM-guided feature selection, leveraging LLM semantic domain knowledge alongside data-driven metrics to identify meaningful, actionable portfolio segments. Evaluated on a mortgage portfolio with 200+ features, Oculi demonstrates effectiveness in discovering material risk segments previously intractable through manual exploration, reducing time-to-insight significantly while maintaining auditability and statistical rigor.

Hidden-state probes effectively detect LLM hallucinations, but the geometry of the signal remains poorly characterized, driving increasingly complex probe architectures. Across three 7B-scale models and three datasets in a paired-example paradigm, we find the signal overwhelmingly dominated by a single mean-shift component, and removing this direction collapses detection to chance. Shrinkage linear discriminant analysis closes about 73% of the gap between 1D and full-dimensional classifiers, so apparent architectural complexity largely reflects high-dimensional covariance estimation difficulty rather than exploitable non-linearity. A simple L2-regularized logistic regression (0.952 AUROC) bounds or outperforms twelve controlled architectural alternatives, and our multi-layer aggregation exceeds CLAP cross-layer attention probing under matched paradigm. Because the signal spans a contiguous layer band, LayerMix aggregates it to match oracle-layer performance without oracle access. Our claims characterize the geometry within the controlled paired-example paradigm. Our code is available at https://github.com/js-lee-AI/LayerMix.

Data augmentation is a cornerstone of deep learning pipelines, yet existing strategies treat it as a static, model-agnostic preprocessing step, either relying on expensive dataset-specific policy search or applying transformations uniformly at random, regardless of what the model has already learned. We introduce ActiveAugment, a unified framework that treats augmentation selection as an online active learning problem. For each training minibatch, ActiveAugment generates a pool of candidate augmented views and scores each candidate using a combination of the model's predictive uncertainty and the feature discrepancy induced by the augmentation. The augmentation under which the current model is most fragile is selected per sample, and the model is then trained with a joint supervised classification and supervised contrastive objective that enforces intra-class invariance to the selected augmentations while maintaining inter-class separation. We evaluate ActiveAugment on eight benchmark datasets spanning natural and medical imaging, using CNN and transformer architectures across three training regimes (training from scratch, full fine-tuning, and linear probing), and comparing eight active selection strategies for augmentation scoring. ActiveAugment outperforms AutoAugment, RandAugment, and TrivialAugment under controlled augmentation shifts across all domains and budgets, with the most pronounced gains at low labelling budgets. On medical imaging datasets, where data is scarce and domain shift relative to natural-image pretrained models is large, ActiveAugment achieves higher test F1 than all baselines, demonstrating strong cross-domain adaptability. Our analysis reveals that the augmentation selection policy evolves meaningfully during training and that strategy choice has a direct impact on generalisation. Code is available at: https://github.com/noahvide/ActiveAugment.