arXiv:2607.11796v1 Announce Type: new
Abstract: Selective state-space models such as Mamba route information through a bank of first-order modes whose input coupling is set by a learned selection mechanism. We give an exact instrument for measuring how a trained model uses these modes. Because the state matrix is diagonal, each channel's output decomposes exactly into per-mode contributions, and a per-(layer, channel, window) Gram tensor yields the exact output error of dropping any subset of modes, offline, at any budget. Validated against the reference implementation to a relative error of $2.3\times10^{-7}$ on the Mamba-1 family where it is exact, the instrument predicts a layer's deployed pruning error to a median relative deviation of $5\times10^{-7}$ over $4{,}464$ configurations, its floor set by the reconstruction. Applying the instrument across the Mamba-1 family (130M--2.8B), the deployed 7B Falcon-Mamba, and Mamba-2, we find that trained models re-allocate their state space with the input: which modes carry the signal migrates across contexts, and at the most affected layers a per-input oracle roughly halves the output error of a fixed mode set. Frozen-signal counterfactuals attribute the migration primarily to the input-dependent write map $B_t$; the timestep usually identified with selectivity carries almost none of it. Input-scheduled mode pruning on this measurement outperforms static, Hankel-based, and layer-adaptive rankings at every scale from 130M to the deployed 7B Falcon-Mamba, and at half the state budget it matches the unpruned model. Because the scheduler reads each window's mode usage from a first pass, this demonstrates realizable headroom; we claim no deployed compute or memory saving.
Science Journals
arXiv:2607.11582v1 Announce Type: new
Abstract: We present Parcas, a concurrent separation logic for verifying the parallel time complexity of fork-join programs. In order to abstract from the specifics of the machine, time complexity for parallel programs is given in terms of two metrics: the work, measuring the total number of operations, and the span, measuring the longest chain of sequential dependencies. Together, these two metrics determine the running time on any number of processors. For proving bounds on the work and span, Parcas is equipped with work credits and span credits, logical devices that represent permissions to incur costs.
Work credits are a straightforward adaptation of time credits, a standard tool for bounding time complexity of sequential programs, and can be split additively between parallel tasks. Span credits, however, require a fundamentally different treatment. Indeed, the span of the parallel composition of two tasks is the maximum of the span of the two tasks. To account for this, we propose a rule for duplicating span credits at fork points, with each copy tagged by a logical task identifier that restricts which task may spend them. A transfer rule allows unused span credits to be forwarded across sequential compositions to subsequent tasks. The logic is expressive enough to give modular, higher-order specifications for common parallel primitives such as a parallel for loop and a tabulate function. We demonstrate Parcas on several case studies, including parallel prefix sums, parallel merge sort, and a variant of Treiber's lock-free stack that mixes concurrency with parallelism. All the presented results are mechanized in the Rocq prover using the Iris separation logic framework.
arXiv:2607.10601v1 Announce Type: new
Abstract: Large Language Model (LLM) agents are commonly trained from expert trajectories using supervised fine-tuning (SFT), which treats multi-turn agent behavior as ordinary text imitation. This recipe is simple and low-cost, but it only learns to imitate the sequence of expert actions, rather than training the agent to choose the right action against plausible mistakes at each state. Existing methods to mitigate this problem include preference learning or reinforcement learning, but they usually need high-cost environment rollouts and reward models. We propose Agentic-DPO, a lightweight offline agent policy optimization method that turns expert trajectories into state-conditioned preference supervision. At each expert action state, Agentic-DPO samples a one-step action from the current state, treats plausible wrong actions as negatives, and contrasts them with the expert action using a DPO-style preference objective. To avoid mixing both policy and schema in preference learning, we introduce Policy-Preserving Augmentation (PPA), which renders the same latent trajectory under multiple schemas while keeping the expert policy fixed. Agentic-DPO requires no online environment rollout, reward model, or full-trajectory student exploration. We conduct experiments across StableToolBench, tau-bench retail, and Mind2Web, where Agentic-DPO consistently improves agents at different model scales beyond imitation. In particular, it raises tau-bench accuracy from 21.7% (SFT) to 41.4% for a 9B model, matching online GRPO under the same backbone with only step-level rollouts and without environment interaction during gradient steps. The results suggest that expert trajectories can support low-cost agentic policy optimization when converted from demonstrations into state-level action preferences. Code for Agentic-DPO is released at https://github.com/Schuture/Agentic-DPO.
arXiv:2607.10313v1 Announce Type: new
Abstract: Dyadic conversational motion generation is essential for realistic interactive digital humans. Existing approaches typically model conversational behaviors within unified dyadic generators. However, such holistic formulations tend to couple self-speech-driven motion with partner-responsive social feedback, leaving the interaction-specific component implicit and underutilizing the speech-motion correspondence already learned by pretrained monologic motion models. We propose Learn2Chat, a unified framework that models dyadic motion as interaction modulation over pretrained monologic motion priors. This design separates intrinsic speech-driven motion from social interaction effects and enables more structured interaction modeling. Specifically, we introduce a Monologic-Anchored Motion Factorization scheme that leverages the semantic motion manifold learned from monologic data to disentangle audio-driven motion dynamics from interaction-induced modulation, yielding clean interaction representations from dyadic sequences. On top of this representation space, a Cross-Attentive Interaction Latent Prediction module maps paired speech signals to interaction latents through cross-branch attention and interaction alignment. During inference, the predicted interaction latents modulate canonical monologic motion to generate coherent and synchronized dyadic behaviors in a data-efficient manner. Extensive experiments on the DualTalk benchmark demonstrate that Learn2Chat achieves state-of-the-art performance across both quantitative metrics and perceptual evaluations. Moreover, the framework is model-agnostic and seamlessly integrates with diverse pretrained monologic motion backbones, highlighting the effectiveness of prior reuse and interaction adaptation for scalable conversational motion generation. More visual results are available on the project page.
arXiv:2607.11803v1 Announce Type: new
Abstract: Climate change is a source of anxiety about the future. Understanding how people express themselves about climate change enables us to address such concerns. To study climate change expression on social media, we analyzed 200 TikTok videos tagged with #climatechange, identifying four categories of content: expression-feelings, views-appeals, news-information, and trend-hijacking. We found that creators use humor to package sharp critiques, avoiding direct confrontation. They replace complex discussions with life stories, such as adopting a vegetarian lifestyle or deleting emails. They borrow from news media to present fragmented information as scientific interpretations, creating a perception of scientific credibility, balancing scientific accuracy with emotionality. Analysis of viewer responses showed they engaged empathetically, reshaping interpretations of videos. These interactions risk reinforcing existing views but help build community on TikTok, which lacks community structure. This study reveals how creators may retell news on science using personal narratives, highlighting how short-form videos enable climate communication.
arXiv:2607.09691v1 Announce Type: new
Abstract: A modern coding agent can hold an entire repository in its context window. Most of its reading is wasted -- and the interesting question is not how much context an agent can use, but what it actually \emph{needs}. We study that question at the moment it matters most: when the agent must \emph{edit} code. Separating \emph{finding} the work site from \emph{acting} on it, we hold localization fixed with an oracle, vary only how the code is represented, and score context against real issue resolution on SWE-bench Verified. The answer is starkly minimal. The signal lives in the code being edited itself: natural-language summaries of it answer almost none of the behavioral questions that the source answers ($4/45$ vs.\ $27/45$, held-out repositories, independent judge), and the gap belongs to the representation, not the summarizer -- a frontier model's summaries score exactly as poorly as a 3B model's. The surrounding context hardly matters either: across every multi-file instance in Verified, under a protocol frozen before any data, rendering a file's remainder as UML skeletons and signatures resolves no more issues than deleting that remainder outright ($N{=}70$, exact McNemar $p{=}0.75$). That was our registered hypothesis, and it failed. Compressed context, meanwhile, matches whole files at a third of the tokens: a resolved issue costs $19$K context tokens, not $94$K. The instrument also yielded a finding the field should keep: temperature-0 API inference flips ${\sim}9\%$ of per-instance outcomes between byte-identical runs. That is a noise floor under every small effect reported on this benchmark, including ours. We release the instrument -- gold-validated environments, per-instance proof that every reference edit is expressible from every arm's context, deterministic patch construction, and pre-registered hypotheses whose nulls we publish.
arXiv:2607.11387v1 Announce Type: new
Abstract: Voice assistants are widely deployed on mobile platforms, yet most are designed as system-level services that remain poorly aligned with application-specific behavior. As a result, enabling voice interaction at the app level requires developers to manually reimplement application logic, leading to high development and maintenance costs.
We propose an LLM-driven approach to automating the development of app-specific voice assistants by repurposing GUI test code, which encodes behavior-preserving, executable specifications of application functionality. In this paper, we present a perspective in which large language models reinterpret GUI tests as bridges between application behavior and conversational interaction. By transforming test methods into app-specific VA artifacts, such as voice intents, capability descriptions, and executable action plans, our approach grounds voice assistants directly in existing application logic rather than external specifications.
We illustrate this vision through AppVA, a research prototype on Android. Our preliminary results across five open-source applications suggest that GUI test code can be reused beyond testing, enabling the synthesis of app-specific voice assistants and highlighting a broader research direction at the intersection of software testing, interaction design, and LLM-enabled automation.
arXiv:2607.09686v1 Announce Type: new
Abstract: Sparse Mixture-of-Experts (MoE) language models separate total parameter count from per-token active computation, but local inference systems often still require the full model, key-value cache, runtime buffers, and operatingsystem headroom to fit in fast memory. MawForge tests a different systems hypothesis: local MoE serving can be made practical on constrained unified-memory machines by storing the full model on disk, keeping common tensors resident, and materializing routed expert tensors into a bounded execution cache on demand. The central finding is that MawForge is effective as a bounded execution mechanism and measurement substrate for local MoE inference, but not as a cache-maximization policy. Performance depends on balancing expert reuse against resident footprint, KV-cache size, quantization, route locality, and macOS memory pressure.
arXiv:2607.09715v1 Announce Type: new
Abstract: This article proposes a model, based on graph theory, to represent a variety of two-player games of perfect information, such as chess and checkers. I then provide a backtracking minimax algorithm to find, if it exists, a perfect game strategy (game resolution), and subsequently a variant of that algorithm which is required to determine a weaker condition: the existence of a strategy to never lose (always reach at least a draw).
Of course, this does not mean that in practice the algorithm can find such a strategy in a short time, but here we are only concerned with formally proving that this is possible, at least theoretically.
arXiv:2607.11207v1 Announce Type: new
Abstract: Table-based reasoning with large language models (LLMs), which requires reasoning based on natural language questions and structured tabular data, has gained widespread attention. However, a series of issues still constrain the application of this task. The previous approaches suffered from significant performance degradation when faced with large tables due to the difficulty of long text modeling and the limitation of input length for LLMs. The text-to-SQL approach is used to efficiently extract key information from tables and generate smaller sub-tables. However, tabular data, especially web tables, often lack the necessary structure and consistency, making them unsuitable for performing mathematical logic operations using SQL queries. We propose the ProgramTab framework, which guides LLMs employing in-context learning to perform tabular data preprocessing with Python code, as well as the momentous contents extraction with row and column extraction and SQL generation. The experiment results on table reasoning datasets demonstrate that the ProgramTab framework effectively deals with table-based reasoning tasks and outperforms all LLM-based baselines.
arXiv:2607.10494v1 Announce Type: new
Abstract: Software engineering increasingly relies on asynchronous communication artifacts, including recorded meetings where stakeholders discuss concerns, rationale, and decisions. These meetings often include diagram-based representations of requirements, system behavior, component interactions, and trace dependencies. Accessing knowledge from these meetings is challenging because recordings are long and relevant evidence is distributed across speech, slides, and technical diagrams. This paper reports our industrial experience developing and evaluating LMVQA, an LLM-based multimodal question-answering system for technical meeting videos. Developed in collaboration with engineers at Ciena, LMVQA supports the understanding of requirements and design intent by grounding answers in audio and visual evidence, with explicit handling of diagram-rich content such as requirements and UML diagrams. It processes each video once to build a reusable time-stamped evidence corpus for grounded question answering. Across a Ciena dataset and a public dataset, we show that LMVQA significantly improves answer accuracy compared to a state-of-the-art baseline, from 31% to 94% on the Ciena dataset and from 21% to 88% on the public dataset, with larger gains on diagram-rich videos. We further show that, after one-time indexing, LMVQA reduces average response time from 81.3s to 3.3s on Ciena and from 98.4s to 9.2s on the public dataset, while lowering average token-based LLM API cost by about 75%. Finally, our interviews with three domain experts show that engineers particularly value LMVQA for locating software-engineering-relevant information, revisiting rationale, and tracing answers to specific video segments.
arXiv:2607.10652v1 Announce Type: new
Abstract: This paper presents a symplectic Hamiltonian direct discontinuous Galerkin (DDG) method for approximating wave propagation problems, including the linear and semilinear wave equations. Within an auxiliary-variable-free DG framework, we prove that the symmetry of the numerical flux bilinear form is equivalent to the existence of a discrete Hamiltonian structure. It follows that methods such as the symmetric interior penalty method and the symmetric DDG (SDDG) method admit a discrete Hamiltonian structure, whereas schemes including the Baumann--Oden, DDG, and BR2 methods do not possess this property. Exploiting this structure, we construct fully discrete symplectic schemes by combining the SDDG spatial discretization with symplectic time integrators. We further derive error estimates for the SDDG method applied to semilinear wave equations, showing the optimal convergence rate for the displacement and the suboptimal convergence rate for the velocity. Numerical experiments validate the theoretical convergence rates and demonstrate that the symplectic Hamiltonian DDG method achieves superior long-time energy conservation and accuracy.
arXiv:2607.10499v1 Announce Type: new
Abstract: Automated verifiers for IEC 61131-3 ladder diagrams enhance safety by translating diagrams into model-checker inputs. Still, their unverified front-end translations risk silently returning incorrect results (missing violations or raising false alarms) when they diverge from the standard. We address this gap with K-ESBMC, an executable formal semantics of IEC 61131-3 ladder diagrams built in the K framework. K-ESBMC models contacts, coils, timers, counters, edge blocks, and the retentive scan cycle, generating both an interpreter and a deductive verifier from a single definition. Validated scan-for-scan against OpenPLC/Matiec, K-ESBMC serves as an independent reference oracle to test the ESBMC Programmable Logic Controller (PLC) ladder diagram to GOTO translation differentially. It agrees with ESBMC on most programs and reproduces injected violations with concrete witnesses. Every disagreement exposes a genuine ESBMC defect, confirmed by OpenPLC and two other verifiers, revealing two failure modes: an unsound skip that certifies unsafe programs, and an imprecise havoc that produces spurious counterexamples. For the combinational and latch fragment, we machine-check in kprove that Kes BMC's rules implement the standard's input/output relation, elevating the correctness argument from empirical to formal. K-ESBMC provides a reusable, standard-faithful oracle for auditing any ladder diagram verifier's translation, offering a general approach to verifying the soundness of translation-based verification tools.
arXiv:2607.10809v1 Announce Type: new
Abstract: A compact and high order HWENO scheme using ADER (Arbitrary high order using DERivatives) time discretization is developed for hyperbolic conservation laws on the triangular mesh, which is the extension of the work on the structured mesh (Luo et. al. (2024) \cite{luo2023}). The Lax-Wendroff procedure is employed to convert time derivatives to spatial derivatives. Thanks to this, the cell averages of the derivatives of the solution can be obtained by the time accurate solution as Gaussian points along the cell interfaces through the Green-Gauss theorem instead of by the evolution solution directly in the conventional HWENO methods. Comparing with the existing Runge-Kutta HWENO (RK-HWENO) method on the unstructured mesh (Zhao et. al. (2025) \cite{zhao2025}), the new method has the following advantages. Firstly, the RK-HWENO method must solve the additional equations for reconstructions and time advancing, which is avoided for the new method. Secondly, the HWENO reconstruction in the new method is performed once per time step and is different from the RK-HWENO method, in which the reconstruction is performed several times every time step. Because of these advantages the new method is more efficient than the RK-HWENO method with smaller numerical errors and less computational costs. Besides, comparing with the existing ADER-WENO methods \cite{dumbser20071,dumbser20072} under the same order of accuracy, the stencil of the new method is more compact since the both the function and its first derivative values are used in the reconstruction of the HWENO schemes. Numerical examples demonstrate that the new method can achieve the high order for smooth solutions both in space and time, keep non-oscillatory near discontinuities.
arXiv:2607.09934v1 Announce Type: new
Abstract: A multi-species Bhatnagar-Gross-Krook (BGK) model for gas mixtures is presented that achieves the correct species-wise relaxation of velocities, temperatures, and pressure tensors according to the Boltzmann collision integral, as well as the correct mixture Prandtl number, while retaining a single relaxation term per species. The model extends the ellipsoidal statistical BGK (ESBGK) model by introducing relative relaxation targets for each species, derived from the Variable Hard Sphere (VHS) production rates of the Grad 13 approximation. Three approaches for the species relaxation frequency are proposed and analyzed: a Grad 13-based per-species frequency, a mixture-averaged frequency, and an empirical harmonic mean of the two. The model is implemented in the particle-based code PICLas and verified against Direct Simulation Monte Carlo (DSMC) results for a range of test cases, including 0D reservoir relaxation, mass diffusion, supersonic Couette flow, and hypersonic flow around a 70{\deg} blunted cone for binary and ternary gas mixtures. Across all test cases, the proposed model reproduces the correct Prandtl number, species temperature, velocity relaxation rates and pressure tensor relaxation, with the empirical relaxation frequency consistently yielding the best agreement with DSMC.
arXiv:2607.10577v1 Announce Type: new
Abstract: Perhaps one of the most intriguing phenomena in time-varying-media photonics is the amplification of light in a photonic time crystal (PTC). However, studies to date have focused only on the PTC-based amplification of coherent light. In this work, we theoretically examine the PTC-based amplification of thermal radiation, specifically blackbody radiation. Such amplification is fundamentally intriguing because of the inherently stochastic nature of thermal radiation, and technologically relevant because of its ubiquity. For simplicity, and because of the experimental relevance of transmission lines, we consider a one-dimensional medium. To analyze the PTC-based amplification of blackbody radiation, we examine the spatial correlations and spatial spectra of the electromagnetic fields. We show that the initially blackbody radiation periodically converges to Gaussian spatial correlations and spectra, with gradually increasing amplitudes, coherence lengths, and both spatial- and wavenumber-domain purities. We further demonstrate that these asymptotic behaviors are governed by the momentum band structure of the PTC and can be understood using a rotating-wave approximation for the pseudo-Hermitian dynamics of an electromagnetic field in a PTC.
arXiv:2607.10077v1 Announce Type: new
Abstract: Tabular learning is still dominated by gradient-boosted decision trees (GBDTs), while recent deep learning approaches have become increasingly competitive. However, applying deep tabular models to large-scale datasets remains challenging, as large sample sizes, high feature dimensionality, or many target classes can introduce substantial computational cost. We propose TabLoRA, a parameter-efficient trainable neural ensemble for large-scale tabular learning. Instead of using fully independent ensemble backbones, TabLoRA shares a common backbone across predictors and introduces predictor-specific low-rank adaptations, enabling ensemble-style prediction without full parameter duplication. Across benchmarks, TabLoRA achieves a favorable balance between predictive performance and practical efficiency compared with GBDT methods and recent deep learning baselines under the same resource constraints. Memory analysis and ablation studies further show that the proposed design improves the feasibility of neural ensemble learning while preserving much of the benefit of full ensembles.
arXiv:2607.10081v1 Announce Type: new
Abstract: The means to execute and orchestrate software components has changed from human-written code to descriptive prose. In high performance computing, this transition is represented in application orchestration, workload management, and system monitoring and debugging, to name a few. The underlying means to enable descriptive definition of tasks is the use of the Large Language Model with associated tool functions and resources. A combination of a model with access to such resources, modeled in software, encompasses an autonomous framework. As fully automated and agentic frameworks are developed for science, it is important to assess reliability and strategies scoped to specific tasks. In this work, we assess the extent to which an agentic framework can optimize and run an HPC scaling study with a low latency network in Amazon Web Services, accurately transform HPC job specifications between workload managers, and design and run an entire biosciences workflow. We find that the framework completes all three tasks while surfacing task-specific failure modes. In the scaling study, agents deploy and optimize applications but monitor running jobs inefficiently, preferring conservative fixed waits over event subscriptions. In job translation, they convert specifications between Slurm and Flux with high accuracy, with processor-affinity flags the most common error. In the bioscience workflow, the agent reproduces an expert-written variant-calling pipeline almost exactly -- agreeing with the reference call set in 18 of 19 completed runs -- and reaches this result through many distinct yet functionally equivalent workflow implementations. This information is invaluable moving forward to developing multi-cluster setups with scheduling and transformation handled by agents.
arXiv:2607.11821v1 Announce Type: new
Abstract: In multi-objective combinatorial optimization, unsupported non-dominated points typically outnumber supported points and are often significantly more challenging to compute. Recent studies show that extreme supported non-dominated points provide high-quality representations of the non-dominated set for certain binary problems. We demonstrate that this observation does not generalize to capacitated network optimization problems: representation quality decreases with increasing arc capacities, whereas supported non-dominated points consistently provide high-quality representations with respect to several quality indicators. However, supported point sets may still be too large in practical applications, where only a small, fixed number of alternatives is typically desired. Selecting fixed-size representations from the non-dominated set requires its computationally expensive generation and thus diminishes the computational advantages that representations are intended to provide. We therefore suggest the (extreme) supported points as alternative candidate sets in subset selection problems. Our numerical results show that restricting the candidate set to supported non-dominated points yields fixed-size representations of nearly the same quality as those selected from the complete non-dominated set. Overall, supported non-dominated points serve both as high-quality representations and as reasonable candidate sets for subset selection.
arXiv:2607.10164v1 Announce Type: new
Abstract: Surface tension is central to many two-phase flows, making accurate numerical schemes essential for predicting its effects. The integral formulation introduced by Popinet and Zaleski (1999) provides a natural discretisation that conserves momentum locally and globally and extends directly to variable surface tension, including Marangoni flows. However, to the authors' knowledge, only two-dimensional formulations have been reported, mainly because robust implementation in three dimensions is challenging for interfaces with complex geometries.
This work presents the first three-dimensional integral surface tension scheme, implemented within a sharp front-tracking framework. The method is tested for static and translating spherical droplets, oscillating droplets, thermocapillary motion, and rising bubbles. Results are compared with analytical solutions, experimental data, and established approaches, including the continuous surface force (CSF) and smoothing-based methods.
The proposed scheme produces spurious velocities comparable to CSF, while providing greater accuracy in all other tests. The largest improvements occur for droplets oscillating at low Ohnesorge numbers, variable-surface-tension flows, and strongly deforming rising bubbles. For a thermocapillary-driven droplet, terminal-velocity errors are reduced by up to five orders of magnitude relative to smoothing-based methods. The predicted steady-state shapes of rising bubbles also agree substantially better with experiments, particularly at low Morton numbers.
arXiv:2607.10195v1 Announce Type: new
Abstract: We investigate the problem of computing the distribution function for the shortest and longest path lengths in a directed graph with random edge lengths. Specifically, when these lengths are uniformly distributed, the problem reduces to computing the volume of a polytope defined by the graph structure. We establish that the problem is $\#P$-hard, even under the restricted condition that the random edge lengths are identically and independently distributed (i.i.d.) according to any continuous probability distribution with certain natural conditions, the local uniformity. This hardness result applies broadly: while the uniform distribution provides an essential case for the reduction, other distributions -- such as exponential or normal -- are similarly hard because they contain uniform distributions in every arbitrarily small interval. Furthermore, we show that the problem is contained within $\mathrm{XP}$ with respect to the treewidth $k$ of the underlying undirected graph. For the specific case of i.i.d. uniform edge lengths, we present a novel dynamic programming algorithm that processes a tree decomposition by iteratively performing convolutions to propagate distribution functions. Our approach achieves a time complexity of $n^{O(k^2)}$ for any fixed treewidth $k$.
arXiv:2607.10335v1 Announce Type: new
Abstract: The integration of power converters is profoundly changing the power system dynamics and poses significant challenges for stability analysis. The dynamic interactions between the power grid and the heterogeneous converters are highly complex and difficult to analyze due to the curse of dimensionality. Moreover, system stability varies with the operating points, which are determined by the voltage magnitude, active power, and reactive power of each converter. This further complicates the analysis as it is difficult to enumerate and examine all the possible operating points. To tackle these challenges, this paper proposes a geometric decentralized stability certificate for power electronics (PE)-dominated power systems, which can simultaneously handle heterogeneous power converters and their variable operating points. The certificate can be checked in a decentralized and modular manner, and it is scalable for large-scale power systems. Our approach is developed based on the concept of Davis-Wielandt (DW) shell and its projections, which can effectively visualize the characteristics of high-dimensional complex matrices. We investigate how the projections of the DW shell vary with operating points and how this variation can guide the search for worst-case operating conditions. We further propose an efficient algorithm to compute the stability margin and construct the certified operating regions. The effectiveness of the proposed method is validated through case studies on single-converter and 54-converter wind power systems.
arXiv:2607.10666v1 Announce Type: new
Abstract: Deploying AI-based visual inspection in manufacturing is hard because requirements change often, new defect types appear, and large labeled datasets are rarely available. We propose answer-conditioned chain-of-thought (CoT) distillation for rapidly adapting small vision-language models (VLMs) to new industrial tasks using minimal labeled data. A frontier VLM receives each training image along with its correct label and generates a justified visual explanation. A 3B-parameter model is then fine-tuned on these reasoning-augmented examples via LoRA. By conditioning on correct answers, we ensure all training reasoning is directed toward the correct conclusion, which is critical because frontier models score as low as 24.1% on our hardest task. We validate on four industrial classification tasks spanning three image modalities using only 18 to 30 labeled images per task. Across 4 seeds per task (32 training runs), our method outperforms direct fine-tuning on all 16 seed-task combinations, with mean improvements of +1.7 to +4.4 percentage points. A controlled equal-budget experiment confirms the improvement comes from reasoning quality, not additional training steps. An unconditioned baseline demonstrates that with out answer-conditioning, wrong reasoning degrades performance by 17.8 percentage points. On weld radiograph classification, the fine-tuned 3B model outperforms GPT-4.1 by 10.0pp using just 24 training images.
arXiv:2607.10539v1 Announce Type: new
Abstract: Existing approaches to infer user traits and generate responses consistent with a persona rely on static prompting. They lack calibrated uncertainty, ignore sequential evidence, and drift during long interactions. We present \textbf{AI YOU}, a framework that continually updates a personality profile with 22 dimensions from conversation and embodies it in a personal digital twin. Practically, the system combines prompting, Bayesian updating, and conformal prediction for persona inference. A periodically refreshed memory anchor and cognitive memory with three layers preserve persona consistency over long interactions. Across the main results, AI YOU \emph{(i)} achieves conformal coverage ranging from 0.921 to 0.976, \emph{(ii)} improves uncertainty calibration and reasoning grounded in memory, and \emph{(iii)} enhances persona fidelity over static prompting in role playing over 100 turns while reducing trait drift, for most evaluated backbones under adversarial settings with multiple agents. The prototype \emph{AI YOU Town} initializes an imaginative twin world for future interaction. The online demo is available at \href{https://quinnnnnne-ai-you.hf.space/}{\mbox{\texttt{quinnnnnne-ai-you.hf.space}}}.
arXiv:2607.10674v1 Announce Type: new
Abstract: As AI code tools become integrated into programming environments, students increasingly describe intended behavior in natural language and rely on these tools to generate code, shifting emphasis from code writing to specification. Yet little is known about the comments students write as specifications in AI-assisted programming tasks. We analyze a four-year dataset of undergraduate programming submissions and reflections from tasks in which students wrote comments to guide code generation and refined solutions using test-case feedback. We introduce a taxonomy spanning three dimensions: comment type, code expression level, and code construct. Using automated classification, we examine how these dimensions vary across attempts and how students describe the process in their reflections. Our findings show that students mostly wrote natural-language What comments, shifted toward How comments for more procedural constructs, and focused more on verifying generated code than on repeatedly rewriting comments.