中 EN
登录 注册账户

Academic Intelligence · Curated Daily

探索全球前沿学术脉络

AcademicHub 汇聚顶级期刊与预印本平台的实时文献。定制您的专属科研雷达，利用大语言模型自动生成交叉领域文献分析简报。

登录 注册账户
01.
arXiv (CS.AI) 2026-06-11

PROJECTMEM: A Local-First, Event-Sourced Memory and Judgment Layer for AI Coding Agents

作者:
Ripon Chandra Malo↗Tong Qiu↗

arXiv:2606.12329v1 Announce Type: new Abstract: AI coding assistants now support a growing share of software work, from quick scripts to production applications. Yet these agents remain largely stateless: each new session re-reads project files, re-derives prior decisions, and - most costly - may repeat debugging attempts that already failed. Reconstructing this context can consume an estimated 5,000-20,000 tokens per session; the bottleneck is often not model capability but missing project memory. We present projectmem, an open-source, local-first memory and judgment layer for AI coding agents. projectmem records development as an append-only, plain-text event log of typed events - issues, attempts, fixes, decisions, and notes - and deterministically projects that log into compact, AI-readable summaries served through the Model Context Protocol (MCP). Beyond storage, projectmem adds a deterministic pre-action gate that warns an agent before it repeats a previously failed fix or edits a known-fragile file. We frame this as Memory-as-Governance: memory that does not merely answer the agent but acts on its next action. The system runs fully offline with no telemetry; its immutable log also serves as a provenance trail for reproducible, auditable AI-assisted development. projectmem ships as a three-dependency Python package (14 MCP tools, 19 CLI commands, 37 automated tests) and is evaluated through a two-month self-study across 10 projects comprising 207 logged events. Source code: https://github.com/riponcm/projectmem.

阅读与讨论 → 访问原文 →
02.
arXiv (CS.CL) 2026-06-16

Know Your Limits : On the Faithfulness of LLMs as Solvers and Autoformalizers in Legal Reasoning

作者:
Olivia Peiyu Wang↗Sanna Wong-Toropainen↗Daneshvar Amrollahi↗Ryan Bai↗Tashvi Bansal↗Arush Garg↗Leilani H. Gilpin↗

Large Language Models (LLMs) achieve strong performance on reasoning tasks, but whether this reflects faithful logical inference or heuristic approximation remains unclear. We study this question in legal entailment by comparing three paradigms, including pure LLM classification, LLM-based Formal Reasoning, and solver-based Formal Reasoning using the Z3 SMT solver, on a re-annotated subset of ContractNLI across five LLMs. Our re-annotation reveals a systematic and measurable gap between pragmatic legal interpretation and strict formal entailment, where a substantial proportion of legally sound inferences are not formally grounded without additional unstated assumptions. While introducing formal structure improves accuracy, with LLM-based Formal Reasoning achieving the highest benchmark performance, we show that this gain does not imply faithful reasoning. We identify three recurring failure modes: scope laundering, where LLMs report solver-inconsistent classifications without executing the underlying formal reasoning, producing conclusions that appear logically grounded but are not; implicit constraint blindness, where LLMs overlook logical constraints present in formal representations; and program synthesis failures, where LLMs generate incorrect Z3 code despite structured prompting. Critically, scope laundering persists across all models, raising serious concerns about the faithfulness of LLM-based formal reasoning as a proxy for symbolic execution. These results reveal a fundamental gap between benchmark accuracy and logical faithfulness.

阅读与讨论 → 访问原文 →
03.
medRxiv (Medicine) 2026-06-12

Mathematical analysis of the overall survival after chemoradiotherapy of limited-stage small cell lung cancer and the effect of dose/fractionation

作者:
Bunuel-Muriscot↗Gonzalez-Crespo↗Otero-Casal↗Gomez-Caamano↗Pardo-Montero↗

The purpose of this work is to analyze the 2-year overall survival (OS2y) of limited-stage small cell lung cancer (LS-SCLC) treated with chemoradiotherapy (CRT), aiming at characterizing the response of LS-SCLC, and in particular the /{beta} value and proliferation parameters. Through a systematic analysis of the literature, we collated a dataset containing 57 entries (3363 patients) of response of LS-SCLC treated with CRT. Radiotherapy schedules ranged from hyper- to hypofractionation. Four radiobiological models to describe the OS2y were investigated, with progressive levels of complexity including the effect of radiotherapy, chemotherapy, treatment year and toxicity. The Akaike Information Criterion (AIC) was used to compare models, and the profile likelihood methodology to compute confidence intervals. Model 4, which includes the effect of radiotherapy, chemotherapy, treatment year and dose-dependent toxicity, provided the best fits of the experimental data (lowest AIC value). While being the best model, model 4 still fails to provide a good prediction of the OS2y, in particular failing to predict the survival of the schedules achieving the lower/higher survivals. The radiobiological analysis of the dose-response of LS-SCLC to CRT does not allow to narrowly constrain the value of response parameters. We attribute this limitation to the large heterogeneity of this disease. Nonetheless, our analysis shows a large /{beta} value (>9 Gy, 95% CI), which implies a low fractionation effect in the radiotherapy of LS-SCLC. and an accelerated proliferation of tumor cells, {lambda}' > 1.6 Gy/day (95% CI), after a kick-off time of ~4-5 weeks, which supports the use of accelerated protocols to avoid the effect of tumor proliferation on the clinical outcome.

阅读与讨论 → 访问原文 →
04.
medRxiv (Medicine) 2026-06-16

Ranking-optimized survival models can underperform fixed-horizon clinical prediction: a SUPPORT2 reanalysis of machine learning, attending-physician judgment, and the original SUPPORT model at 60- and 180-day mortality

作者:
Truong↗Q. H↗Hoang↗D. C↗Luu↗D. T↗

Machine-learning survival models are increasingly proposed for intensive-care mortality prediction and are almost always selected and reported using the concordance index, a ranking metric averaged over follow-up. Yet most bedside decisions hinge on a probability at a specific time, such as 60- or 180-day mortality. We asked whether ranking-optimized models remain competitive at fixed clinical horizons against two reference points clinicians actually rely on: unaided attending-physician judgment and the original 1995 SUPPORT logistic model. Reanalyzing the SUPPORT2 cohort (9,105 critically ill adults from five United States centers, 1989-1994) under a stratified 70/15/15 split, we compared a gradient-boosted survival model, the physician's recorded prognosis, and the 1995 model at 60 and 180 days, alongside several alternative learners. The survival model achieved competitive ranking concordance (0.705) yet underperformed both comparators at fixed horizons: at 60 days its area under the ROC curve was 0.750, against 0.808 for physicians on the matched sample and 0.827 for the 1995 model, a gap that held across eight independent data splits and remained statistically reliable after multiplicity correction. The shortfall was not miscalibration, since post-hoc recalibration left discrimination unchanged, nor limited capacity, since neural networks, a deep ranking model, and two timepoint-aware discrete-time models also failed to close it; replacing the ranking objective with timepoint-matched binary training recovered roughly half the gap, pointing to an objective-horizon mismatch. Discrimination was equitable across sex, race, and age, but leave-one-disease-out validation exposed severe failure for disease groups absent from training, and the physician advantage was conditional on a physician electing to provide an estimate. We recommend reporting timepoint-specific discrimination alongside concordance, timepoint-matched training when fixed-horizon predictions drive care, leave-one-subgroup validation, and distribution-free prediction intervals to support selective deployment.

阅读与讨论 → 访问原文 →
05.
arXiv (CS.AI) 2026-06-11

On the Geometry of On-Policy Distillation

作者:
Zhennan Shen↗Yanshu Li↗Qingyu Yin↗Chak Tou Leong↗Zhilin Wang↗Yanxu Chen↗Rongduo Han↗Sunbowen Lee↗Yi R. Fung↗

arXiv:2606.07082v2 Announce Type: replace-cross Abstract: On-policy distillation (OPD) is increasingly used to improve large language model reasoning, but its training dynamics remain poorly understood. We characterize the trajectory of OPD updates in parameter space and compare it with supervised fine-tuning (SFT) and reinforcement learning with verifiable rewards (RLVR). A suite of parameter-space diagnostics consistently places OPD in a relaxed off-principal regime: compared with SFT, its updates affect fewer weights and avoid principal directions more strongly, while compared with RLVR, they remain less tightly constrained. Beyond this static localization, OPD exhibits subspace locking: its cumulative updates rapidly enter a narrow low-dimensional channel. Constraining training to the update subspace formed early in training preserves OPD performance but substantially degrades SFT, indicating that the locked subspace is functionally sufficient for OPD. Control experiments further show that sparsifying the update tokens and shifting rollout generation off-policy preserve the rank dynamics, whereas mixing the OPD objective with RLVR changes them. Overall, these results suggest that OPD is not merely an intermediate point between SFT and RLVR, but induces its own update geometry in parameter space.

阅读与讨论 → 访问原文 →
06.
arXiv (CS.CV) 2026-06-16

SGFormer++: Semantic Graph Transformer for Incremental 3D Scene Graph Generation

作者:
Mengshi Qi↗Changsheng Lv↗Zijian Fu↗Xianlin Zhang↗Huadong Ma↗

In this paper, we propose SGFormer++, a novel Semantic Graph Transformer for 3D scene graph generation (SGG), which aims to parse point cloud scenes into semantic structural graphs, where nodes denote detected object instances and edges encode their pairwise relationships, with the core challenge lying in modeling complex global scene structure. While existing graph convolutional network (GCN)-based methods suffer from over-smoothing and limited receptive fields, SGFormer++ leverages Transformer layers as its backbone to enable global message passing. Specifically, we introduce two key components tailored for 3D SGG: (1) a Graph Embedding Layer++ that efficiently integrates edge-aware global context with linear computational complexity, and (2) a Semantic Injection Layer++ that enriches visual features with linguistic priors from large language models (LLMs) and vision-language models (VLMs), boosting semantic representation without introducing extra trainable parameters. To further address the practical challenge of incremental SGG (I-SGG), where new relationship categories arrive sequentially, we equip SGFormer++ with a novel Spatial-guided Feature Adapter, which calibrates predicate features using subject-object spatial geometry to counter scale variation, and a Cascaded Binary Prediction Head that mitigates catastrophic forgetting via task-incremental classifier expansion and logit distillation. Extensive experiments on the 3DSSG benchmark demonstrate that SGFormer++ achieves state-of-the-art performance in both standard and incremental settings: it yields a significant 4.49% absolute improvement in Predicate A@1 under the incremental setting. Code and data are available at: https://github.com/Andy20178/SGFormer.

阅读与讨论 → 访问原文 →
07.
arXiv (CS.AI) 2026-06-18

Structured Representation Learning with Locally Linear Embeddings and Adaptive Feature Fusion

作者:
Somjit Nath↗Jackson J Cone↗Derek Nowrouzezahrai↗Samira Ebrahimi Kahou↗

arXiv:2606.18469v1 Announce Type: cross Abstract: Neuroscientific research has revealed that the brain encodes complex behaviors by leveraging structured, low-dimensional manifolds and dynamically fusing multiple sources of information through adaptive gating mechanisms. Inspired by these principles, we propose a novel reinforcement learning (RL) framework that encourages the disentanglement of dynamics-specific and reward-specific features, drawing direct parallels to how neural circuits separate and integrate information for efficient decision-making. Our approach leverages locally linear embeddings (LLEs) to capture the intrinsic, locally linear structure inherent in many environments, mirroring the local smoothness observed in neural population activity, while concurrently deriving reward-specific features through the standard RL objective. An attention mechanism, analogous to cortical gating, adaptively fuses these complementary representations on a per-state basis. Experimental results on benchmark tasks demonstrate that our method, grounded in neuroscientific principles, improves learning efficiency and overall performance compared to conventional RL approaches, highlighting the benefits of explicitly modeling local state structures and adaptive feature selection as observed in biological systems.

阅读与讨论 → 访问原文 →
08.
arXiv (CS.CL) 2026-06-11

Substrate Asymmetry in User-Side Memory: A Diagnostic Framework

作者:
Youwang Deng↗

User-side memory in LLMs is typically scored as a single "personalization" capability: given a user's history, is the output more user-aware? We show this aggregate metric hides opposite-direction failures. Memory factorises into at least three orthogonal axes – behavioral consistency (style, voice), factual presence (recall facts in history), and factual absence (abstain when a fact is absent) – and no single substrate wins all three. Comparing per-user gamma-LoRA (a small LoRA adapter trained on each user's history; gamma denotes per-user, not per-task) against BGE-large dense top-K retrieval on a controlled 50-user synthetic corpus and a real-data probe (LaMP-3), we find gamma-LoRA decisively wins behavioral style while RAG decisively wins factual absence – and the same query-projection cells in attention layers 21-35 causally load-bear both effects in opposite directions (zeroing those LoRA weights raises absence-probe TPR by +33 pp and drops presence-probe TPR by 20 pp). On the more heavily RLHF-tuned Llama-3.1-8B-Instruct the asymmetry strengthens, not heals: parametric memory's behavioral advantage collapses while its absence-calibration deficit against retrieval widens – an alignment tax on parametric user-memory. On real-data LaMP-3, gamma-LoRA underperforms a majority baseline; a 9-condition mitigation sweep diagnoses this as instruction-following collapse, not substrate failure (a 9x2 cross-product shows the eval-time {1..5} logit mask drives main_acc to >=0.995 on every recipe), and the best training-time fix replicates bit-identically on Llama. Finally, substrate-selection routing is question-classification, not calibration: a 110M DistilBERT on the question text alone beats every logit-based router. We contribute the diagnostic framework, the diagnosed real-data negative, the alignment-tax replication, and the routing-as-classification finding.

阅读与讨论 → 访问原文 →
09.
arXiv (CS.LG) 2026-06-16

Causal-Privacy Audit Workflow for Synthetic and Distilled Data in Dropout Support

作者:
Hanghang Zheng↗Xiwei Zhuang↗Zhong Wang↗Hong Liu↗Xiao Chen↗Jingwen He↗Xia Li↗

arXiv:2606.15940v1 Announce Type: new Abstract: Synthetic and distilled student data are increasingly used to enable privacy-conscious learning analytics, yet their suitability for decision-facing institutional support remains uncertain. In dropout support, generated data must preserve not only predictive utility or distributional resemblance, but also the financial-status evidence used to guide advising, payment-plan assistance, and scholarship-related decisions. Method: This study introduces CaP-Eval, a decision-facing causal-privacy audit workflow for evaluating generated student data under a fixed estimand, timing-aware adjustment design, estimator set, and empirical privacy-governance screen. The workflow compares original, distilled, adversarial synthetic, statistical synthetic, and DPGNet privacy-oriented generated data on predictive utility, treatment-effect fidelity, robustness to alternative estimators, and local training-record proximity. Results: DPGNet and distilled data preserved the original financial-status treatment-effect structure more reliably than the adversarial and Gaussian Copula baselines. DPGNet preserved full direction and rank agreement across epsilon levels; epsilon = 10 produced the smallest non-original IPW and DML deviations, while epsilon = 1 and epsilon = 5 amplified several financial-status contrasts. Distilled data remained highly faithful but retained the strongest local training-record proximity signal. TabularGNet preserved qualitative directions with moderate attenuation, and Gaussian Copula compressed effect magnitudes. Conclusions: Predictive utility, privacy orientation, empirical disclosure signals, and causal fidelity diverged; generated student data require joint audits of direction, magnitude, overlap, and release-governance risk before decision use.

阅读与讨论 → 访问原文 →
10.
arXiv (quant-ph) 2026-06-12

Accidental Symmetry in the Tavis-Cummings Model via the Schwinger Boson Representation

作者:
Plato Deliyannis↗Iman Marvian↗

arXiv:2606.12813v1 Announce Type: new Abstract: The Jaynes-Cummings (JC) Hamiltonian is a paradigmatic model of light-matter interaction and, more generally, qubit-boson interactions, widely used across atomic, optical, and superconducting qubit platforms. In the multi-qubit setting, where n qubits are identically coupled to a single boson mode, this interaction is known as the Tavis-Cummings (TC) Hamiltonian. The structure of the TC model is usually understood in terms of two standard symmetries: permutation invariance of the qubits and a U(1) symmetry associated with conservation of the total excitation number. Here we identify an additional, independent "accidental" symmetry of the TC Hamiltonian and construct the corresponding conserved observable. We show that, for n>2 qubits, this symmetry imposes strong constraints on the realizable unitary transformations. These constraints persist in the presence of the global $J_z$ Hamiltonian, but are removed by adding $J_z^2$, even though $J_z^2$ preserves both permutation invariance and the U(1) symmetry. Finally, we explain the origin of this previously unnoticed symmetry using Schwinger's boson representation of angular momentum. These restrictions have important implications for controllability of the TC system and for its applications to quantum computing, which are investigated further in a companion paper.

阅读与讨论 → 访问原文 →
11.
PLOS Computational Biology 2026-06-02

A comparative study of simulation-based inference methods for epidemic models with identifiability considerations

作者:
Geunsoo Jang↗

by Geunsoo Jang, K. Selçuk Candan, Gerardo Chowell Epidemic models play a critical role in understanding transmission dynamics, generating forecasts, and informing public health interventions when they are properly calibrated to epidemiological data. Traditional Bayesian inference methods rely on the likelihood function to update prior knowledge using observed data. However, for realistic epidemic models, likelihood functions are often analytically intractable or computationally prohibitive, which can limit the applicability of these methods. Simulation-based inference provides a promising alternative by approximating posterior distributions through forward simulations rather than an explicit likelihood evaluation. In this study, we present a systematic comparison of four approaches: Approximate Bayesian Computation (ABC), Neural Posterior Estimation (NPE), a neural method with temporal embedding, and Preconditioned Neural Posterior Estimation (PNPE), which integrates elements of both classical and neural techniques. These methods are evaluated across epidemic models of increasing complexity under fixed simulation budgets and varying levels of observational noise, with explicit attention to both structural and practical identifiability. Our results show that neural methods generally improve posterior fidelity and predictive accuracy compared with ABC under constrained simulation budgets. PNPE achieved strong performance in several simulation settings, whereas temporal embeddings improved inference in models with complex epidemic dynamics by capturing sequential dependencies. These gains come with important trade-offs: PNPE required substantially greater computational resources and, unlike fully amortized NPE-based methods, may require reconditioning for each new observation. In contrast, ABC remained computationally efficient and provided reasonable, though often more conservative, posterior estimates. Overall, our findings highlight trade-offs among computational efficiency, posterior accuracy, uncertainty calibration, and inference reusability, suggesting that method selection should depend on model complexity, data quality, identifiability, and available computational resources.

阅读与讨论 → 访问原文 →
12.
arXiv (CS.CL) 2026-06-16

Stop When Further Reasoning Won't Help: Attention-State Adaptive Generation in Reasoning Models

作者:
Jiakai Li↗Ke Qin↗Rongzheng Wang↗Yizhuo Ma↗Qizhi Chen↗Muquan Li↗Shuang Liang↗

By incorporating test-time compute scaling, large reasoning models (LRMs) can solve complex problems through explicit chain-of-thought (CoT) reasoning processes. However, they often suffer from overthinking, resulting in redundant token outputs and degraded accuracy. Current methods to mitigate this issue remain limited: training-based approaches require substantial computational resources, while training-free methods rely on well-crafted prompts or unreliable confidence signals. In this work, we investigate early stopping from the perspective of attention distributions and propose a simple method, ASAG, which infers the model's reasoning state and adaptively adjusts the generation strategy. The proposed framework is training-free and plug-and-play, enabling seamless integration into existing LRMs. Extensive experiments on nine benchmarks demonstrate consistent improvements across mainstream LRMs with varying parameter scales, including the DeepSeek-R1-Distill and Qwen3 series. Specifically, ASAG improves average accuracy by 3.2% while reducing the number of generated tokens by nearly 40% across all reasoning tasks on Qwen3-8B.

阅读与讨论 → 访问原文 →
13.
arXiv (math.PR) 2026-06-11

Mean-field theory via dissociated arrays for particle systems interacting through noisy weights

作者:
Nicolas Fournier↗Datong Zhou↗

arXiv:2606.12135v1 Announce Type: new Abstract: We study a mean-field limit for a $N$-particle system in which each particle follows a diffusion and interacts with other particles through a weight on each directed edge. Each weight evolves according to its own nonlinear SDE driven by a Brownian motion, with coefficients involving the states of the two endpoint particles of the edge. The initial vertex and edge variables are assumed to have a dissociated Aldous–Hoover form. We construct the limiting nonlinear SDE by averaging the interaction over an independent neighbor and an edge input, prove its well-posedness, and show that the dissociated vertex-edge structure is propagated by the dynamics. This propagation property is an analogue of propagation of chaos in the case where the weight of each edge may remain correlated with the states of the two endpoint particles. Under either a bounded-observable assumption or a sub-Gaussian edge-input condition, the finite system converges to this limit through quantitative coupling estimates for a typical particle and a typical edge. We also prove the convergence of the empirical measure of particle's state pairs and their interaction weights.

阅读与讨论 → 访问原文 →
14.
arXiv (CS.AI) 2026-06-11

When Context Returns: Toward Robust Internalization in On-Policy Distillation

作者:
Xun Wang↗Ruishuo Chen↗Zhuoran Li↗Yu Chen↗Longbo Huang↗

arXiv:2606.11627v1 Announce Type: cross Abstract: Recent work has shown that on-policy distillation can internalize privileged context, such as system prompts or task hints, into a student model so that the context is no longer needed at inference time. Although this approach successfully improves the student's no-context performance, we identify an interesting and previously unstudied phenomenon: in many settings, reintroducing the original privileged context to the distilled student actually degrades its performance, even on instances it already solves correctly without context. We term this context-induced degradation and argue that robust internalization demands not only matching the teacher's context-conditioned behavior, but also remaining stable when the context is reintroduced, a property we call context removability. Motivated by this observation, we propose a lightweight consistency regularizer that first anchors the student's no-context output via stop-gradient, then penalizes the context-conditioned output for deviating from it via forward KL divergence. This simple addition requires only one extra forward pass per training step, yet it effectively mitigates context-induced degradation and, in many cases, even improves no-context performance. Across 12 configurations spanning diverse domains and model families, our method improves context-conditioned accuracy in the majority of settings, reduces context-induced harm in 11 out of 12 settings, and effectively eliminates response-length inflation. A mechanistic case study further confirms that context removability is achieved at the representation level, with hidden states remaining nearly identical regardless of whether the context is present.

阅读与讨论 → 访问原文 →
15.
arXiv (CS.CL) 2026-06-18

CEO-Bench: Can Agents Play the Long Game?

作者:
Haozhe Chen↗Karthik Narasimhan↗Zhuang Liu↗

Language model agents are becoming proficient executors at isolated, short-horizon tasks such as software engineering and customer service. Yet real-world challenges require a combination of sophisticated skills that remain largely untested in agents: (1) navigating long horizons amid uncertainty; (2) acquiring information in noisy environments; (3) adapting to a changing world; (4) orchestrating multiple moving parts toward a coherent goal. We introduce CEO-Bench, which evaluates these capabilities together by simulating a representative real-world task: operating a startup for 500 days. An agent manages pricing, marketing, budgeting, and many other aspects of a fictional company through a programmable Python interface, operating in the same environment and facing the same challenges as a human CEO. Success demands analyzing noisy, interconnected business databases, translating signals into sound strategy, and coordinating many decisions with programming. The strongest agents write sophisticated code that simulates customer cohorts to forecast future cash and mines negotiation history to uncover hidden customer preferences. Even so, most state-of-the-art models struggle in this environment. Only Claude Opus 4.8 and GPT-5.5 finish above the $1M starting balance, and neither consistently turns a profit. CEO-Bench takes a first step toward measuring the intelligence required to drive sustained, adaptive progress over time.

阅读与讨论 → 访问原文 →
17.
arXiv (CS.AI) 2026-06-11

Market Design for AI: Beyond the Copyright Binary

作者:
Yan Dai↗Maryam Farboodi↗Negin Golrezaei↗Sepehr Shahshahani↗

arXiv:2606.12260v1 Announce Type: cross Abstract: How can we design a market of human-generated content for use in training AI models that both enables technological progress and preserves individual incentives for high-quality content creation? Existing approaches take polar positions: a "free-for-all" model based on fair use and a "strong intellectual property rights" model. We show that both fail: Free-for-all does not compensate creators, and – by modeling as a static Stackelberg game – strong intellectual property rights also underpower creative incentives. We find this especially true for more innovative creators, a phenomenon we term the "originality penalty." Extending this insight to a dynamic model, we find another market failure undermining AI model performance, even for an initially good model: Such a model induces greater reliance by humans on AI-assisted creation, resulting in homogenized content feeding back into training, which degrades the model performance – a "curse of precision." We further propose a market design with a data intermediary internalizing cross-creator externalities and subsidizing innovative contributions, thereby restoring efficiency.

阅读与讨论 → 访问原文 →
18.
arXiv (CS.AI) 2026-06-11

Categorical Prior Lock-in: Why In-Context Learning Fails for Structured Data

作者:
Antonio Pelusi↗Stefano Braghin↗Alberto Trombetta↗

arXiv:2606.11961v1 Announce Type: cross Abstract: Large language models (LLMs) are increasingly used as conditional generators for structured data, relying on in-context learning (ICL) to adapt to new distributions without parameter updates. We investigate the limits of ICL for structured generation under distribution mismatch, using high-cardinality tabular data as a controlled test case, and identify a structural failure mode we term categorical prior lock-in: the inability of ICL to update the model's prior over token distributions inherited from pre-training. Across two 7B-parameter open-weight models, ICL improves numerical fidelity with additional examples but exhibits a sharp ceiling on categorical distributions, failing to reproduce rare classes entirely. Parameter-efficient fine-tuning (LoRA) overcomes these limitations but introduces measurable memorization risk and, in some cases, destabilizes structured output generation, highlighting a fundamental trade-off between adaptability and privacy.

阅读与讨论 → 访问原文 →
19.
arXiv (CS.AI) 2026-06-16

Decision-Weighted Flow Matching for Contextual Stochastic Optimization

作者:
Jize Xie↗Haomiao Wu↗Qiang Chen↗Xiu Su↗Yi Chen↗

arXiv:2606.16790v1 Announce Type: cross Abstract: Conditional generative models are increasingly used as scenario generators for stochastic optimization, but standard training objectives emphasize uniform distributional fit rather than the downstream decisions induced by generated scenarios. This creates an objective mismatch: errors in statistically common regions may have little effect on decision regret, whereas errors in decision-sensitive regions can substantially change the optimal action. We propose Decision-Weighted Flow Matching (DW-FM), a regret-aligned training framework that preserves the simplicity of standard flow matching while reweighting its velocity-regression objective using decision-sensitive endpoint information. Theoretically, we connect downstream regret to pathwise velocity mismatch through a loss-induced decision discrepancy and an adjoint transport argument, yielding an ideal regret-aligned surrogate and practical endpoint-weighted objectives with regret guarantees. Empirically, we demonstrate the effectiveness of DW-FM on three CVaR-based contextual stochastic optimization benchmarks spanning synthetic portfolio, semi-real financial, and traffic-CVaR tasks, where DW-FM improves downstream regret over standard baselines.

阅读与讨论 → 访问原文 →
20.
arXiv (math.PR) 2026-06-16

Balanced affine Motzkin paths: Pearson geometry and global endpoint asymptotics

作者:
Alexander Omelchenko↗

arXiv:2601.17634v2 Announce Type: replace Abstract: We study endpoint distributions of balanced affine weighted Motzkin paths. In the balanced case, the generating-function equation has Pearson-type characteristic geometry. We show that this geometry controls the terminal-height law globally: the characteristic escape time determines the limiting cumulant generating function, the large-deviation rate function, and the ray-scale asymptotics. Thus the usual Gaussian window is only the local quadratic approximation to a global Pearson-driven profile. For finite sizes, we prove a uniform Daniels saddlepoint approximation in the one-dominant-singularity regimes and identify the exceptional antipodal case requiring a lattice/interference correction.

阅读与讨论 → 访问原文 →
21.
arXiv (CS.CL) 2026-06-16

Building Customer Support AI Agents at 100M-User Scale: An Evaluation-Driven Framework

作者:
Aman Gupta↗Kevin Rossell↗Edesio Alcoba\c{c}a↗Jose Chrystian Lima Pacheco↗Carolina Baptista de Lima↗Shao Tang↗Luiz Paulo Rabachini↗Luis Moneda↗Herbert Fei↗Daniel Silva↗Rohan Ramanath↗

The rapid rise in LLM capabilities has made AI agents increasingly viable across a broad range of tasks. Among the most promising applications is building production-ready customer-facing agents, a challenge that demands coordinated excellence in evaluation methodology, context engineering, training, and online measurement. Yet these critical pillars are typically developed in isolation, creating blind spots that only surface after deployment. In this paper, we present a unified framework that bridges offline development with online impact for customer support AI agents at Nubank, a company with 100M+ users. Our approach integrates several key components: (1) structured context engineering tailored to customer support agents, (2) systematic human-in-the-loop prompt iteration, (3) rigorous LLM judge evaluation with measured inter-rater agreement and GEPA optimization for consistency, and (4) ideation-to-production validation. A central insight is that evaluation-pipeline quality directly determines iteration velocity. We present results from five production deployments spanning distinct domains: card delivery, debt management, credit-limit support, card management, and product explanation. These deployments deliver consistent customer-satisfaction gains while substantially accelerating iteration. In our card-delivery deployment, large-scale A/B testing yields a 37 percentage-point improvement in AI transactional Net Promoter Score and a 29 percentage-point gain in self-service rate over prior agent variants, alongside a strong correlation between offline simulation metrics and online outcomes, demonstrating that eval-driven development reliably predicts production impact. On most use cases, AI satisfaction reaches within a few percentage points of expert human agents.

阅读与讨论 → 访问原文 →
22.
arXiv (CS.LG) 2026-06-11

Capacity-Constrained Online Convex Optimization with Delayed Feedback

作者:
Alexander Ryabchenko↗Idan Attias↗Daniel M. Roy↗

arXiv:2606.11711v1 Announce Type: new Abstract: Online learning with delayed feedback typically assumes that the learner can track all pending rounds until their feedback arrives. In practice, tracking resources are finite, and feedback from untracked rounds is permanently lost. In this paper, we study delayed online convex optimization (OCO) under a hard capacity constraint, where at most $C$ pending rounds can be tracked at any time. To model delay information, we introduce a semi-clairvoyant model that refines the clairvoyant assumption from prior work: rather than requiring delays to be known at prediction time, the learner observes delay expirations online, consistent with the classical unconstrained delayed setting. Our approach proceeds via a reduction to a novel ``delayed and weighted'' OCO problem, using a scheduler that randomizes tracking decisions and importance-weights the resulting observations. For this base problem, we propose and analyze Delayed-Weighted FTRL and its bandit analogue, establishing regret bounds that explicitly characterize the interaction between time-varying weights and delayed feedback. Combining these base learners with our schedulers yields the first regret guarantees for capacity-constrained OCO under convex and strongly convex losses, for both first-order and bandit feedback. For first-order feedback, capacity $C = \Omega(\log T)$ suffices to recover standard delayed OCO rates up to logarithmic factors. For bandit feedback, the regret rates are modulated by powers of $(1 + \sigma_{max}/C)$, where $\sigma_{max}$ is the maximum number of pending observations at any time. This allows the regret bound to degrade gracefully when $C < \sigma_{max}$, while remaining sublinear.

阅读与讨论 → 访问原文 →
23.
arXiv (CS.AI) 2026-06-16

Rethinking Scaffolding in LLM Tutors: The Interactional Mismatch Between Benchmarks and Real-World Deployments

作者:
Alexandra Neagu↗Jeffrey T. H. Wong↗Marcus Messer↗Rhodri Nelson↗Peter B. Johnson↗

arXiv:2606.15766v1 Announce Type: new Abstract: A central pedagogical value evaluated in AI tutor benchmarks is scaffolding: guiding students through graduated steps toward a solution. Alignment and evaluation methods for embedding scaffolding behaviour into chatbots, however, rest on an implicit assumption: that students will take up the scaffolding and engage in the conversation. To examine whether this assumption holds, we introduce an evaluation pipeline around two metrics - Chatbot Scaffolding and Student Uptake - and apply them across nine datasets of 9,490 chats, spanning AI tutor benchmarks and real-world deployments of educational chatbots. Our analysis reveals that while benchmarks assume a high-scaffolding, high-student-uptake environment, students in real-world settings exhibit lower levels of uptake overall - frequently bypassing the chatbot's pedagogical framing to drive the interaction toward their own learning goals at little interpersonal cost. We argue that bypassing scaffolding is not necessarily detrimental; rather, it frequently highlights a mismatch between a chatbot's pedagogical framing and the student's learning goals. To meaningfully evaluate the effectiveness of a chatbot's assistance, future benchmarks must move beyond the assumption that students will simply take up the scaffolding, and instead evaluate how these chatbots navigate diverse learning contexts and student-driven interaction patterns.

阅读与讨论 → 访问原文 →
24.
arXiv (CS.AI) 2026-06-12

Evaluation Sovereignty in Metadata-Driven Classification: A Multi-Track Framework for Weakly Supervised Information Systems

作者:
Raymond Vasquez↗

arXiv:2606.13436v1 Announce Type: new Abstract: Evaluation in machine learning is typically treated as a neutral measurement process. However, in operational information systems, evaluation outcomes are often conditioned by the processes used to generate labels. This paper does not seek to improve classification performance. Instead, it examines the validity of performance measurement under differing label-authority regimes. This issue is particularly relevant in large-scale metadata-driven systems, where labels are often incomplete, inconsistent, or weakly supervised. We introduce evaluation sovereignty, defined as the degree to which performance metrics are independent of label authority and supervision regime, and propose a multi-track evaluation framework that systematically varies training and evaluation label sources. Using hierarchical multi-label classification on large-scale scientific metadata, we demonstrate that models exhibiting strong performance under operational ("silver") evaluation degrade substantially under independent ("gold") evaluation, particularly for fine-grained classification. For example, Micro-F1 decreases from approximately 0.54 to 0.03. Notably, ranking-based metrics remain above baseline, revealing a divergence between latent model signal and classification validity. These findings suggest that commonly reported performance metrics may reflect alignment with labeling processes rather than true predictive capability. We therefore reconceptualize evaluation validity as a system-level property shaped by label governance and provide a practical methodology for auditing intelligent systems operating under weak supervision.

阅读与讨论 → 访问原文 →
25.
arXiv (CS.AI) 2026-06-16

Green SARC: Predictive Cost and Carbon Governance for Agentic AI Systems

作者:
Gaston Besanson↗

arXiv:2606.15954v1 Announce Type: cross Abstract: Agentic AI systems act through tools and sub-agents, yet the controls meant to bound their financial and environmental cost still sit on dashboards evaluated beside or after execution. Green SARC applies the SARC governance-by-architecture framework – four enforcement sites in the agent loop – to FinOps and GreenOps, contributing the theory of what to enforce and how to predict it. We report four policy-independent results. (i) The unconstrained "State Snowball" is $\Theta(n^2)$ in loop depth; on 3,000 real multi-step plans (SWE-rebench) it holds on 100%, with median curvature $\hat{c}_2=216$ exceeding the linear-accretion prediction $p/2=134$ – real plans accrete faster than the model. (ii) On real residuals the Normal-$\sigma$ gate under-covers (92% at nominal 95%); split-conformal calibration holds (95.2%). (iii) A soft Lagrangian penalty tuned to the budget in expectation breaches it on 91.5% of seeds; the architectural gate breaches 0%. (iv) Under binding budgets the gate's over-budget incidence is 0% on synthetic and real (BurstGPT) arrivals. End-to-end token/USD/carbon savings (47–55%) are real but policy-dependent in magnitude – set by a scope-cap knob, not by gate rejections. The library is open-source, dependency-free, and ships a regeneration script for every cited number.

阅读与讨论 → 访问原文 →