Peer Revenue Equity in science

Papers

Fresh from arXiv, typeset properly. Highlighted words trade on the market.

All of arXiv

Friday 28 August

Artificial Intelligence

AcCoRD: Evaluating User-Agent Collaboration Under Realistic User Preference Dynamics

User preferences in user-agent collaboration are rarely static and fully-specified upfront: preferences are formed, revealed, adjusted, and relaxed during interaction. Existing benchmarks for evaluating user-agent collaboration focus almost exclusively on resolving underspecified preferences, thereby failing to capture the richer dynamics of real-world interaction. We introduce AcCoRD, a user-agent collaboration benchmark requiring agents to handle diverse user preference dynamics in two domains: online shopping and travel planning. We evaluate five frontier LLMs under two prompting strategies: vanilla ReAct, and an uncertainty-guided variant that prompts models to identify and resolve ambiguity about user preferences. Our results reveal that frontier models can handle underspecification but struggle to satisfy preferences that emerge or evolve mid-interaction and require more sophisticated uncertainty modeling. Further, prompting alone fails to elicit the required uncertainty recognition. We release AcCoRD as a resource for developing agents that can navigate the full complexity of real-world user preferences.
Read the full abstract

All authors Tejas Srinivasan, Shikib Mehri, Nandita Shankar Naik, Anirban Das, William M. Campbell, Jesse Thomason

Sound

Auditing generative audio calls for known-task audio-llm evaluation

Speech and audio LLMs are evaluated by comparing waveform predictions with predictions from an automatic speech recognition (ASR) transcript. For fixed closed-set tasks, this conflates acoustic evidence with the need to invoke a generative audio model. We estimate incremental call value with matched selectors sharing pre-call evidence. Each policy may retain the transcript label, use a local encoder, or invoke a generative model; matched control removes generative actions but preserves pre-call evidence and development selection. On VocalSound, transcript-only accuracy is 0.296, while supervised CLAP and WavLM controls reach 0.850 and 0.854 without calls. Full selector reaches 0.925 at 12.5% calls versus 0.921 for matched No-call selector (difference 0.004; 95% CI [-0.025, 0.033]). Thus, results do not show a call gain after transcript and encoder evidence are available. Relevant quantity is incremental accuracy from allowing calls, not the waveform-transcript gap.
Read the full abstract

Computation and Language

PersonaEdit: Representative Sample Selection for Personalized Model Editing

Personalization has attracted growing interest in LLM applications, yet existing retrieval-based approaches depend heavily on retrieval quality and degrade in long-term interactions. Model editing, which directly modifies internal model parameters to incorporate new knowledge, has demonstrated effective knowledge modification capabilities in factual knowledge editing tasks and may provide a potential solution for personalization. However, scaling model editing to personalization is non-trivial. Editing large amounts of user data increases computational cost and causes interference among edits, motivating the need for effective sample selection. To address this issue, we propose, PersonaEdit, a hidden representation clustering strategy that selects representative editing samples through proportional stratified sampling. Experiments show that model editing is effective for personalization, and that our selection strategy preserves most of the performance while substantially reducing the number of required editing samples. Beyond standalone editing, we find that combining model editing with retrieval-based prompt augmentation further improves personalization, as edited knowledge and retrieved context provide complementary information. These results demonstrate the potential of model editing as an efficient and scalable approach for LLM personalization.
Read the full abstract

Analysis of PDEs

The asymptotic structure of forward scattering

Perturbed plane waves are fundamental objects in scattering theory on Euclidean space and asymptotically Euclidean spaces. In this paper, we investigate the structure of perturbed plane waves in the $\textit{forward}$ direction, in which the outgoing spherical wave is typically singular and conjoined to the incoming plane wave. Melrose & Zworski provided a microlocal description (in the more general setting of asymptotically conic manifolds) using their notion of Lagrangian distributions associated to pairs of intersecting Legendrian submanifolds, on the way to proving that the S-matrix is an FIO. Here, we revisit the problem in the asymptotically Euclidean case, for which the oscillatory integrals used by Melrose–Zworski attain their most complicated form (relative to the more general asymptotically conic case). We seek a more elementary description in terms of physical-space asymptotics. These are specified using a two-faced compactification $X\hookleftarrow \mathbb{R}^d$, with one face for each asymptotic regime. We prove full polyhomogeneity. A transport equation arises as a model problem at the main face (`bf'). The quantum inverted harmonic oscillator arises as a model problem at the front face (`ff').
Read the full abstract

All authors Nicholas Lohr, Izak Oltman, Ethan Sussman, Yuzhou Joey Zou

Instrumentation and Detectors

Migration of Belle II TOP Feature Extraction from the Zynq Processing System to PCIe40 Readout PCs

The Belle II Time-of-Propagation (TOP) detector is a key system for charged-particle identification (PID). Although TOP operated successfully during initial Belle II data taking, increasing luminosity and beam-induced background led to more frequent single-event upsets (SEUs) in the radiation-exposed Zynq systems-on-chip (SoCs). Resulting lockups of the embedded processing systems (PSs) interrupted data acquisition. To mitigate this limitation, waveform feature extraction, a critical task of the on-detector Zynq PS, was migrated to the off-detector PCIe40 readout PCs (ROPCs). The new architecture bypasses the PS in the event data path and performs feature extraction outside the detector radiation environment. This change removed SEU-induced PS lockups from the event data path and enabled stable operation of the TOP front-end electronics and the Belle II data-acquisition system under increased luminosity and background conditions. The typical TOP deadtime decreased from about 1% to a level consistent with zero after the final firmware patch, and stable operation was demonstrated at a Level-1 (L1) trigger rate of 30 kHz with a microchannel-plate photomultiplier-tube (MCP-PMT) hit rate of approximately 5 MHz per PMT. This paper describes the migrated architecture, its deployment and validation, and its operational and PID performance.
Read the full abstract

All authors Harsh Purwar, Shahab Kohani, Vasily Shebalin, Martin Bessner, Matt Andrew, Thomas Browder, Tommy Lam, Kurtis Nishimura, Sourav Patra, Vladimir Savinov, Marko Staric, Gary Varner, Satoru Yamada, Keisuke Yoshihara

Computation and Language

Representation of syntax in LLMs through the lens of linear distance and similarity-aware entropy

Structural probes were introduced by Hewitt and Manning to reconstruct syntactic trees from a neural language model's latent representations. They are evaluated by calculating the proportion of syntactic tree edges correctly reconstructed over an annotated corpus (as measured by undirected unlabeled attachment score). Here, we disaggregate this measure, considering undirected attachment score by label (UASL), which assesses the reconstruction accuracy of each syntactic relation separately, establishing important differences among relations that overlap linguistic distinctions. Moreover, we identify two factors that predict most of UASL's variability across relations: (i) the mean and dispersion of the linear distance (on a log scale) between the related words, and (ii) the diversity (similarity-aware entropy) of the syntactic relation's head. These results, which hold across a range of model sizes and architectures, shed light on the degree of abstraction of the representation of syntax in language models and the dependence of such representation on geometric properties of the embedding space.
Read the full abstract

All authors Juan Pablo Vigneaux, Mary Kennedy, Khalil Iskarous, Robert Frank, Matilde Marcolli

Materials Science

Compositionally Complex Ceramics (CCCs): A New Paradigm in Ceramics

The development of high-entropy ceramics (HECs) over the past decade has extended the exploration to a diverse range of oxides, borides, silicides, carbides, nitrides, fluorides, and other ceramic solid solutions, encompassing an increasingly broad range of crystal structures and bonding characteristics and exhibiting promising mechanical, thermal, and functional properties. Initial studies predominantly focused on five-component equimolar compositions. In 2020, we proposed extending the exploration beyond HECs to the broader class of "compositionally complex ceramics" (CCCs), in which rational non-equimolar compositional designs and long- and short-range order can reduce configurational entropy while providing additional opportunities to tailor and enhance materials properties, with the goal of outperforming their higher-entropy counterparts. Because CCCs have been defined somewhat vaguely in previous publications, this review seeks to establish more rigorous and explicit criteria for defining CCCs, despite the unavoidable degree of arbitrariness in the choice of thresholds. Meanwhile, 10-21-component ultrahigh-entropy ceramics have emerged as a subset of HECs, exhibiting intriguing abrupt phase transitions. Dual-phase CCCs have also been reported, with thermodynamic equilibria governing cation partitioning between the two phases. Recent studies have further revealed that grain-boundary complexion (phase-like) transitions can occur in CCCs, influencing microstructural evolution and materials properties. Overall, CCCs provide a versatile platform for tailoring materials properties through the interplay of diverse crystal structures and bonding characteristics, compositional complexity, rational non-equimolar designs, long- and short-range order, defects, and microstructural and interfacial engineering.
Read the full abstract

Geometric Topology

Uncomputable trisection invariants of 4-manifolds

We prove that two invariants of smooth 4-manifolds defined in terms of trisections are uncomputable: the trisection genus and the Kirby-Thompson L-invariant. That is, there does not exist an algorithm that takes as input a triangulated closed orientable 4-manifold and outputs one of these quantities. The result on the L-invariant resolves the second part of Problem 4.116 in the K3 problem list. In the same spirit, we show that the PL multisection genus of PL manifolds is uncomputable in dimensions at least four.
Read the full abstract

All authors Nathan M. Dunfield, Marc Kegel, Shana Yunsheng Li, Qiuyu Ren

Algebraic Geometry

Metric-geometric Chow theorem

In 2009, Peterzil and Starchenko proved the following beautiful generalization of Chow's theorem: An entire complex analytic set $X\subset \mathbb{C}^n$ that is definable in an o-minimal structure on $\mathbb{R}$ must be an algebraic set. This result is known nowadays as the o-minimal Chow theorem. In this article, we present some geometric and metric versions of Chow's theorem that generalize the o-minimal Chow's theorem. For instance, we prove that if $X\subset \mathbb{C}^n$ is a pure $d$-dimensional entire complex analytic set, then the following statements are equivalent: (1) $X$ is a complex algebraic set; (2) $\mathcal{H}^{2d+1}(C(X,\infty))=0$, where $\mathcal{H}^{k}(A)$ denotes the $k$-dimensional Hausdorff measure of $A$, and $C(X,\infty)$ denotes the tangent cone at infinity of $X$; (3) $\mathcal{H}^{2d+2}(C_{\mathbb{C}}(X,\infty))=0$, where $C_{\mathbb{C}}(X,\infty)$ denotes the complex tangent cone at infinity of $X$; (4) $\mathcal{H}^{2d}(Z(X,\infty))=0$, where for $A\subset \mathbb{C}^k$, $Z(A,\infty)=\{[v]\in \mathbb{C}P^{k-1};v\in C_{\mathbb{C}}(A,\infty)\}$; (5) For any $k\in\{d+1,...,n\}$ and for any projection $\pi\colon \mathbb{C}^{n}\to \mathbb{C}^{k}$ such that $\pi^{-1}(0)\cap C_{\mathbb{C}}(X,\infty)=\{0\}$ and $Y=\pi(X)$, $\mathcal{H}^{2d}(Z(Y,\infty))<\mathcal{H}^{2d}(\mathbb{C}P^{d})$.
Read the full abstract

Information Retrieval

LINE Conversation History Retrieval for Personal Memory RAG: Evaluating Search Representations and Hybrid Retrieval

As an initial step toward personal memory retrieval-augmented generation (RAG) for large language models (LLMs), this study presents a retrieval-only case study over one user's LINE conversation history. We segmented 358,896 messages into 22,329 temporally coherent chunks and constructed three search representations: raw_text, a generated summary, and embedding_text, which combines a summary with a raw-text excerpt and other fixed text. We compared BM25, dense vector retrieval, and linear hybrid retrieval on 100 evaluation questions verified by a single annotator. Among individual retrievers, embedding_text_bm25 achieved the highest point estimate, with Recall@5 of 0.584. We then explored six retriever pairings and 21 weights, for 126 configurations on the same evaluation set. The selected combination of embedding_text_bm25 and embedding_text_vector at beta = 0.45 achieved Recall@5 = 0.697, MRR@5 = 0.595, and nDCG@5 = 0.575. Its Recall@5 exceeded that of embedding_text_bm25 by 0.113, with a question-level paired percentile-bootstrap 95% confidence interval of [0.048, 0.184]. This interval is conditional on fixing the configuration selected on the same 100 questions and does not account for uncertainty from configuration selection or weight search. The difference from a summary-based hybrid at beta = 0.50 was 0.050, with a 95% confidence interval of [-0.013, 0.115], so no clear difference could be established. The 17 aggregate questions also yielded lower point estimates than the other question types, suggesting that flat chunk-level retrieval struggles when evidence is distributed across multiple times and conversations. This evaluation is an exploratory single-user, single-annotator study conducted on the same question set used for configuration search; it does not evaluate final answer generation or generalization to unseen questions.
Read the full abstract

Artificial Intelligence

CURA: Certified Runtime Alarms for Computer-Use Agents

Self-report is the cheapest oversight channel a deployer has, and on capable computer-use agents (CUAs) it fails precisely where oversight matters. On 361 OSWorld tasks our pipeline, a read-only feasibility gate, a planner, and a GUI executor, reaches a mean task score of 82.9, above the 72.4 human reference, yet 64 of its 71 failures (90%) end with a success claim, 61 acknowledging no blocker, and the explicit failure affordance is never used in roughly 9,100 calls. We introduce CURA (Certified Runtime Alarms for Computer-Use Agents), an external monitor that reads only harness-visible telemetry, with no model internals, extra LLM calls, or prompt changes, and turns the running trajectory into a sequential test with certified false-alarm control. At alpha = 0.10 its CUSUM alarm detects 42.3% of failures a median of 31 steps before termination at a realized false-alarm rate of 0.066, and risk is partly resolvable before the first action (gate probe, 0.69 AUROC). Retrospectively the composite reaches 0.828 AUROC (fold-internal floor 0.802), but its margin over a total-token baseline is not significant (Delta = +0.026, p = 0.101); the separation is online, where CURA recalls more at matched certified budgets: 0.41 versus 0.34 at alpha = 0.10, 0.56 versus 0.38 at alpha = 0.20. Alarm-gated mid-execution oversight recovers 23 of 70 failures while spending a frontier overseer on 38, giving a deployable cascade at mean score 86.8 and 84.5% full-solve (305 of 361). The certificate bounds false alarms only. We also report where behavioral monitoring is uninformative.
Read the full abstract

All authors Divake Kumar, Sina Tayebati, Devashri Naik, Amanda Sofie Rios, Nilesh Ahuja, Omesh Tickoo, Ranganath Krishnan, Amit Ranjan Trivedi

High Energy Physics (Phenomenology)

Future of Artificial Intelligence for Science in Japan 2024 Community Report

This white paper summarizes scientific challenges and AI/ML research opportunities identified through the FAIRS Japan 2024 unconference process. The discussion focuses on three major physics domains: accelerator physics, cosmology and astrophysics, and neutrino physics. Although each domain has distinct scientific goals and experimental constraints, several common technical themes emerge: high-dimensional reconstruction, fast and accurate simulation, uncertainty propagation, simulation-to-data mismatch, anomaly detection, real-time decision-making, and shared infrastructure.
Read the full abstract

All authors Yoshitaka Itow, Jia Liu, Hirokazu Maesaka, Vinicius Mikuni, Nhat-Minh Nguyen, Hironao Miyatake, Atsushi J. Nishizawa, Patrick de Perio, Daniel Ratner, Kazuhiro Terao, Leander Thiele, Omar Alterkait, Francois Drielsma, Rocio Garcia, Masako Iwasaki, Ahsani Hafizhu Shali, Federica Tarsitano, Takahiro Terada, Junjie Xia

Chemical Physics

OrbGNN: A Wave function-based Machine Learning Interelectronic Representation

Machine learning interatomic potentials (MLIPs) have become emerging tools in molecular modeling and computational chemistry. By learning high-dimensional potential energy surfaces from quantum chemical data, MLIPs enable accurate and efficient predictions of structural, thermodynamic, and dynamical properties. However, such models have limitations in predictions of electronic properties and the effects of static electron correlation due to their lack of electronic structure information. This work presents OrbGNN, an electronic structure graph architecture analogous to molecular graph and MLIP frameworks, where pair-orbital interactions constitute the graph representation, while orbital entanglement encodes the connectivity between them. By embedding information derived from orbital correlation metrics directly into the graph topology, OrbGNN provides a compact representation of a molecule s orbital landscape and electron correlation patterns. Analysis of the behavior of the feature space in an orbital graph are shown to demonstrate model robustness. The model is evaluated for the dissociation of nitrogen and for a larger dataset of diatomic molecules. Finally, the OrbGNN model is applied to a set of octahedral iron(II) complexes to predict spin-state energy gaps.
Read the full abstract

All authors Brody Quebedeaux, Shahzad Akram, Markus Reiher, Konstantinos D. Vogiatzis

Data Structures and Algorithms

The Power of Local Marginals: An $O(\varepsilon^{-1})$-Aspect-Ratio Reduction for Dynamic Weighted Matching

We study dynamic maximum weight matching under edge insertions and deletions in two settings: maintaining a $(1\pm\varepsilon)$-approximation to the optimum weight, and maintaining an explicit $(1-\varepsilon)$-approximate matching. Our main result is a reduction that transforms instances of polynomial aspect ratio into instances of aspect ratio $O(\varepsilon^{-1})$. The reduction applies to general graphs in both settings and is compatible with partially dynamic updates. The reduction is based on a structural property of local marginals. After grouping edges into weight classes, the global marginal contribution of one class relative to all lower classes is approximated by its marginal contribution within a local weight window of aspect ratio $O(\varepsilon^{-1})$. Summing these local marginals yields a value composition lemma that approximates the optimum weight in the entire graph with approximate optimum weights of the local windows. This improves the value reduction of Gupta and Peng (FOCS 2013), whose local aspect ratio is $\varepsilon^{-\Theta(\varepsilon^{-1})}$. The same structural property yields an improved matching composition lemma for explicit matchings, reducing the local aspect ratio of Bernstein–Chen–Dudeja–Langley–Sidford–Tu (SODA 2025) from $O(\varepsilon^{-2})$ to $O(\varepsilon^{-1})$.
Read the full abstract

Combinatorics

Grundy Total Domination and Skew Zero Forcing in Cartesian Products of Paths and Cycles

We determine the Grundy total domination number and the skew zero forcing number for every Cartesian product of two paths or cycles, thereby closing the nonmatching bounds previously known for these rectangular, cylindrical, and toroidal families. For $2\le a\le b$, \[ Z_-(P_a\square P_b) =a-\mathbf1_{\{a\ {\rm odd},\,b\ {\rm even}\}}. \] For $p\ge2$ and $c\ge3$, \[ Z_-(P_p\square C_c)= \begin{cases} \min\{p,c\},&c\text{ is odd}, \min\{2p,c\},&c\text{ is even}. \end{cases} \] Finally, for $3\le a\le b$, \[ Z_-(C_a\square C_b)= \begin{cases} 2a-1,&a=b\text{ odd}, 2a,&a=b\text{ even}, \min\{b,2a\},&a<b,\ a\text{ odd}, a,&a<b,\ a\text{ even},\ b\text{ odd}, 2a,&a<b,\ a,b\text{ even}. \end{cases} \] In every case, $\gamma_{\mathrm{gr}}^t(G)=|V(G)|-Z_-(G)$. The path-containing lower bounds use Kronecker differences of skew-symmetric or hollow symmetric factor matrices, and they also determine maximum skew nullity and minimum skew rank. They further produce an infinite family for which maximum skew nullity is strictly smaller than skew zero forcing. The new cycle–cycle lower bounds use a cyclic column-defect estimate, applied to complete columns when the shorter cycle is odd and, after a one-sided bipartite reduction, to half-columns when both cycles are even. Explicit forcing constructions give matching upper bounds throughout.
Read the full abstract

High Energy Physics (Theory)

Seiberg dualities for quiver gauge theories

We study Seiberg duality for quiver gauge theories with fundamental, bifundamental, and rank-two tensor matter. The existence of a Seiberg dual description places strong constraints on the spectrum of undressed mesons, which is naturally encoded in a factorization equation for the matrix Hilbert series. We derive this factorization from a closed-form large-$N$ superconformal index for general quiver gauge theories with these matter contents. We apply the method to two-node quivers and recover known $SU$–$SU$ and $SO$–$USp$ product-group dualities. We also discuss finite-$N$ consistency conditions from anomaly and operator-spectrum matching.
Read the full abstract

Number Theory

The Multiplicative Persistence Conjecture: Resolving the \(2\)-Adic Obstruction for Nonzero Even Targets

The multiplicative persistence problem studies the process of repeatedly replacing a positive integer by the product of its digits until a single digit, called the terminal digit, is reached. The classical conjecture asserts that no decimal integer requires more than \(11\) iterations. Brier, Clavier, Gutsche, and Naccache proved the conjecture for all odd terminal digits. In their approach to nonzero even terminal digits, they were led to infinite families of decimal integers in which the numbers of digits \(2,\ldots,9\) are fixed, while arbitrarily many digits \(1\) may be inserted. They conjectured that, despite the infinitude of such a family, the exponent of \(2\) dividing its elements is uniformly bounded. We prove this conjecture and obtain an explicit bound depending only on the prescribed digit multiplicities. More generally, our argument applies in every base \(b\geq3\) and to every prime \(p\mid b\). In base \(10\), combining our bound with the method of Brier, Clavier, Gutsche, and Naccache yields a finite algorithm for a further analysis of each nonzero even terminal digit.
Read the full abstract

Analysis of PDEs

A center manifold reduction approach to the Darcy-Bénard convection problem with non-zero Prandtl number

We study the bifurcation of two-dimensional Darcy-Bénard convection (DBC) in a rectangular domain, a canonical model for thermal convection in porous media with applications in geophysics and engineering. The momentum equation lacks advection and viscous dissipation, being regularized solely by a linear Darcy damping term. As a result, the linearized operator generates a semigroup that is neither analytic nor compact and the nonlinear term fails to be Lipschitz. The system is thus placed outside the scope of the standard center-manifold theorem. To overcome these obstructions, we develop a center-manifold reduction adapted to DBC system. Our main result is a constructive proof of the existence of the center manifold function h and local attractivity of the center manifold-the exponential convergence of small solutions toward it-without relying on analyticity of the full linear semigroup and Lipschitz nonlinearity. We circumvent these difficulties by exploiting the partially dissipative structure: the temperature equation is governed by the Laplacian, which generates an analytic semigroup and provides the smoothing needed to compensate for the lack of regularity in the velocity and the absence of global Lipschitz bounds. Through carefully designed inequalities, we establish both the construction of the center manifold function h and the exponential convergence of nearby solutions. Explicit approximations for the center manifold are derived in two scenarios-one simple eigenvalue and two distinct eigenvalues-yielding reduced systems of ordinary differential equations whose analysis determines the bifurcation type. Numerical simulations are presented to corroborate the theoretical results.
Read the full abstract

Cryptography and Security

ContextLeak: Exfiltrating LLM Agent Context via Malicious Tools

Exfiltrating an LLM agent's runtime context – such as the user prompt, execution trajectory, and tool list – poses severe security and privacy risks to users. Such attacks can be carried out via malicious tools and typically require three conditions: (1) the agent selects the malicious tool for task execution, (2) the agent passes its runtime context as input arguments to the tool, and (3) the tool's implementation transmits these inputs to an attacker-controlled endpoint. Existing work primarily focuses on conditions (1) and (3), leaving condition (2) largely unexplored, despite its critical role in enabling successful context exfiltration. In this work, we bridge this gap by developing ContextLeak, a malicious tool attack that induces the agent to both select the tool and disclose its context as input arguments. We realize this attack by carefully crafting the tool's name and description using reinforcement learning. Specifically, ContextLeak employs an LLM, referred to as the attack LLM, to automatically generate the malicious tool's name and description. To improve attack effectiveness, we fine-tune the attack LLM via reinforcement learning on a set of shadow users with diverse, simulated agent contexts. Our key technical contribution is the design of novel reward functions tailored to the context exfiltration objective, enabling effective reinforcement-learning-based fine-tuning of the attack LLM. Extensive evaluation demonstrates that our attack remains highly effective even when the shadow users' contexts differ substantially from those of the victim users. Moreover, ContextLeak significantly outperforms existing malicious tool attacks when adapted to this setting.
Read the full abstract

All authors Yuqi Jia, Ruiqi Wang, Patrick Li, Yuepeng Hu, Peinian Li, Neil Gong

Human-Computer Interaction

How Much Can AI Understand? Toward AI-Assisted Sensemaking of Collaborative Discussion in Groups with Shared History

AI tools that support collaborative discussion typically treat the discussion as a standalone task, focusing only on its content and setting aside the social context of the group having it. But it is groups with a shared history, with their own norms, hierarchies, and relationships, where the most tangled and complex discussions tend to arise. These discussions cannot be understood apart from that context, and AI that overlooks it risks failing to convey what a discussion means, or even misrepresenting it. Drawing on two studies of how experienced Wikipedia editors read and make sense of discussions, we propose an AI-Assisted Sensemaking Model for Collaborative Discussions, which captures not only a discussion's arguments but also the norms and participants behind it, along with the context that gives each meaning. In this model, the system supports the early stages of the sensemaking process, and the degree to which it performs interpretive work can range from low to high. We argue that higher interpretive work reduces the burden on users but increases their reliance on the system's judgment. We then discuss the risks of an insufficiently intelligible system, what it would take to make one more intelligible, and the safeguards it still requires.
Read the full abstract

Programming Languages

Cambria: Resource Abstraction for Parametrized Algebraic Effects and Handlers

The algebraic effects and handlers paradigm separates the concerns of the interface and implementation of computational effects in programming languages. We present Cambria, a language that extends this framework to the parametrized setting. Effect signatures may use abstract parameter types that are instantiated by the handler along with the operation implementations. Parameters abstract over resources, such as memory locations or thread IDs, permitting algebraic effects to encode dynamic allocation. They are first-class in the type system but erased at runtime, requiring no coercions or type-directed reduction. We prove parametricity via a step-indexed logical relation, formalizing the abstraction guarantee provided by parametrized handlers. We also establish type safety and classify the annotations needed for completeness of the type inference algorithm. We demonstrate Cambria's practicality with a working implementation and provide examples including local state, Pólya's urn, and concurrent thread management. The last is a parametrized effect whose abstract thread IDs are shared between concurrent computations, going beyond standard instances. Cambria is the first calculus with user-defined resource-allocating effects that guarantees, via parametricity, that client code cannot depend on how a handler represents its resources.
Read the full abstract

Artificial Intelligence

CEDAR: Automata as Verifiable Interfaces for Language-Guided Embodied Action

Natural-language tasking of embodied agents is rarely just goal specification: users also impose constraints that must persist while the world changes. Code-generating LLM agents can produce plausible behaviors for such instructions, but their free-form programs provide no stable object to verify, compose with new constraints, or repair from a failing trace. We present CEDAR, a counterexample-guided framework that grounds instructions as regular languages over environment event traces. CEDAR uses a language model for semantic judgments and execution traces for correction, then represents both skills and specifications as deterministic finite automata. This turns constraints into executable finite-state objects: a learned skill can be intersected with a learned sleep at night or stay in this biome specification, yielding a controller that enforces the learned constraint by construction rather than by repeated prompting. In Minecraft, with the same simulator/API observations available to a program-generating baseline, CEDAR maintains temporal and spatial constraints that the baseline fails to preserve and amortizes reuse of learned skills, reducing cumulative LLM queries. These results suggest that regular languages offer a practical verification layer between natural-language instructions and embodied-agent policies.
Read the full abstract

Artificial Intelligence

ReToolSQL: Agentic Reinforcement Learning for Robust Text-to-SQL

Recent work has shown that reinforcement learning from execution feedback can substantially improve text-to-SQL performance, often enabling smaller models to match or exceed much larger systems. However, most existing approaches treat SQL generation as a single-turn task, limiting the model's ability to recover from errors through iterative refinement. We present ReToolSQL, a two-stage training framework for text-to-SQL that combines (i) a supervised warm-start on rejection-sampled reasoning traces with (ii) agentic reinforcement fine-tuning (RFT) over multi-turn tool-use trajectories. The key insight is that the two stages act on complementary axes, the supervised fine-tuning (SFT) on verified privileged-teacher traces expands the set of solvable questions (raising pass@k coverage on the hardest cases), while RFT converts that expanded capability into higher single-pass accuracy by teaching the model when to verify, what evidence to retrieve, and how to repair faulty SQL from execution feedback. Applied to Gemma 4 instruction-tuned (31B), RFT alone achieves 73.66% execution accuracy (EX) on the BIRD-SQL development benchmark (74.12% EX with self-consistency). Initializing RFT from the SFT checkpoint (SFT$\to$RFT) yields our strongest model at 74.32% EX single-pass and 74.77% EX with self-consistency. At the time of writing, this ranked first on the BIRD single-model development-set leaderboard. The approach uses composite rewards anchored on execution correctness, requires no human annotation beyond the benchmark itself, and operates within a single dense 31B model, showing that a properly designed SFT$\to$RFT pipeline over tool-use trajectories is a practical path toward robust enterprise-grade text-to-SQL.
Read the full abstract

All authors Pratik Kakkar, Chandra Dhir, Ravi Shankar, Pareekshit Reddy Gaddam, Anup Shirgaonkar

Computer Vision

uScenes: A Multimodal RGB and 3D Sonar Dataset for Underwater Robot Perception

Robust perception is essential for the deployment of autonomous underwater robots. However, optical cameras become unreliable under poor illumination and backscatter. Forward looking (2D) acoustic sensors remain effective under these conditions, but they measure range and bearing while leaving elevation unresolved, creating an ambiguity that prevents individual sonar returns from being localized in three dimensional (3D) space. This complicates the sensor use for 3D scene understanding and precise object detection. We introduce uScenes, a multimodal underwater dataset containing synchronized 3D multibeam sonar point clouds and RGB imagery. The dataset contains 110 scenes and 95,834 synchronized observation, representing 277.6 minutes of data collected across multiple field sessions. uScenes establishes a foundation for underwater sensor fusion, cross modal representation learning and 3D scene understanding. Code and datasets are given at https://github.com/era-research-lab/uScenes.
Read the full abstract

All authors Trung Tien Dong, Zhenqi Wu, Aditya Penumarti, Zi-Hao Zhang, Micaiah Bartlett, Jane Shin, Xiaomin Lin

Machine Learning

Node-wise Feature Encoding for Neural Performance Prediction

As neural networks are increasingly deployed on resource constrained edge devices, accurate prediction of latency and energy is critical for efficient neural architecture search. Existing GNN and transformer based predictors achieve strong results but largely ignore node-level computational cost, limiting their ability to model performance critical operations. To address this, we introduce FeatureFormer, a neural performance predictor that incorporates explicit node-wise encodings of FLOPs, parameter counts, and memory proxies within a gated graph attention architecture. We also present NNEQ, a new large-scale energy consumption dataset that enables unified evaluation of latency and energy prediction. Extensive experiments demonstrate that FeatureFormer achieves state-of-the-art performance across both metrics, including challenging out-of-domain settings. Finally, we show that the proposed encoding is broadly applicable and consistently improves existing predictors with negligible overhead.
Read the full abstract

All authors Matthew Grenier, William Hammer, Andrew Heuer, Nikhil Krishna, Yi Wang, Ramtin Zand

Robotics

CAVE-NAV: VLM-Based Autonomous 3D Navigation in Underwater Cave Environments

Autonomous navigation in underwater cave environments is essential for search-and-rescue operations, scientific exploration, and emergency egress. Traditional navigation systems commonly depend on dense visual features for localization and mapping. In underwater caves, however, visual degradation can undermine feature-based localization, sonar-based mapping may yield overly conservative obstacle representations, and communication constraints preclude real-time human guidance. To address these limitations, we propose an autonomous underwater cave navigation framework that leverages a vision-language model (VLM) with Chain-of-Thought (CoT) reasoning to infer navigable directions from environmental cues, including light intensity gradients, passage morphology, and geometric complexity, captured through multimodal inputs comprising RGB imagery, depth maps, and sonar-based vertical-clearance measurements, thereby supporting safe 3D navigation through confined cave passages. High-fidelity simulations across multiple cave topologies demonstrate that the proposed framework completes all evaluated end-to-end traversals without collisions while maintaining safe clearance from cave boundaries.
Read the full abstract

All authors Zhenqi Wu, Yuanjie Lu, Yisheng Zhang, Miao Yu, Xuesu Xiao, Jaejeong Shin, Xiaomin Lin

Solar and Stellar Astrophysics

Quiescent and Interacting Compact Remnant Binary Populations in the Space Telescope Era

Since the discovery of Gaia BH1, interest in quiescent binary systems has grown. Two additional black-hole binaries have since been discovered alongside several candidate neutron-star binaries, yet the formation pathways of these systems remain poorly understood. Theoretical studies of these systems, together with the populations into which they evolve, such as X-ray binaries (XRBs), can help constrain their formation channels. We present predictions for the Galactic population of quiescent binary systems and their later evolutionary phases, including XRBs and post-interaction quiescent binaries. We combine stellar models from the Binary Population and Spectral Synthesis code (BPASS) with a synthetic Milky Way generated using Feedback In Realistic Environments (FIRE) simulations. We compare our results with existing observations to assess how well the synthetic population reproduces the observations. We find generally good agreement, with observed donor masses, compact-object masses, and orbital periods largely reproduced, except for the Gaia BH systems. Our results also suggest that some accreting compact remnants with low-mass donors and wide orbital periods may not be X-ray bright. Remaining discrepancies indicate that the BPASS initial parameter space should be expanded, the common-envelope evolution prescription revised, and magnetic braking included in low-mass donor models. We also predict the number of systems detectable in future Gaia data releases and by the Nancy Grace Roman Space Telescope (hereafter referred to as Roman). In particular, the Galactic Bulge Time Domain Survey is expected to yield a significant population of quiescent black-hole binaries. We predict that ~200-600 candidate quiescent binary systems will be detected in the Galactic Bulge Time Domain Survey, while ~2500-4000 candidate systems will be identified in Gaia DR5.
Read the full abstract

Machine Learning

Initialization Is Critical: Advancing Federated Short-Term Load Forecasting under Load Heterogeneity via Model Initialization

Short-term load forecasting (STLF) provides essential information for numerous applications in modern power systems. However, accurate STLF often relies on fine-grained smart-meter data from distributed users, raising increasing concerns about data privacy. Federated learning (FL) has therefore emerged as a promising privacy-preserving paradigm for STLF. Nevertheless, this paper reveals structured heterogeneity in clients' load data. Specifically, clients exhibit different responses to exogenous factors and distinct temporal load profiles, which can degrade forecasting performance in FL. To mitigate these issues, this paper studies the role of model initialization in federated STLF, and proposes two initialization strategies from global and local perspectives. For global model initialization, when auxiliary public load data are available, a pretrained initialization strategy is developed to initialize the global model before federated training, thereby reducing client drift during the training process. For local model initialization, we propose SLIAvg, a sequential local initialization strategy that promotes a more consistent training process by allowing participating clients to start from progressively adapted models within each communication round. Since the proposed strategies only modify the initialization process, they are compatible with most existing FL frameworks and privacy-enhancing techniques. Experiments on real smart-meter data with two representative forecasting architectures demonstrate that the proposed strategies effectively improve forecasting performance, as evidenced by reduced client drift, improved convergence behavior, and lower forecasting errors.
Read the full abstract

All authors Jianing Chen, Vajiheh Farhadi, Yan Li, Thomas La Porta

Artificial Intelligence

Credo: Reusable Declarative Primitives for Agentic Workflows

An LLM application depends on both a model and a harness: the program that determines what each call sees, how many calls to make, and which answers to trust. Coding agents can now discover strong harnesses by searching over candidate programs, but the resulting artifact is an opaque block of imperative code whose logical steps, runtime signals, physical execution decisions, and prompt strategies remain implicit and task-specific, forcing subsequent tasks to start the harness search process from scratch. The potential for reuse, however, is substantial. A searched harness encodes significant knowledge, such as the logical steps that work, the signals that matter, the physical operator decisions that adapt execution, and the prompt strategies that are effective, yet this knowledge is buried in imperative code with no inspectable or reusable structure, nor does it carry any provenance or metadata. Credo addresses this problem by recovering a structured declarative description of a searched harness, tagging each extracted primitive with relevant metadata, and cataloguing all of it with provenance. A compiler can then bind stored primitives to generate harnesses for new tasks without having to start the search over from scratch. This paper provides preliminary results demonstrating the potential of our approach and lays out a related research agenda that the database community is well-positioned to tackle, including cost-based compilation over declarative catalogs and catalog maintenance under model and workload drift.
Read the full abstract

Thursday 27 August

High Energy Physics (Theory)

Commuting integrals of motion for free bosonic CFTs revisited: certified densities, spectral rigidity, and an exact identification at the c=1 branch point

We revisit the commuting local integrals of motion for free bosonic CFTs constructed in hep-th/0404195 as input data for integrable boundary states, using exact-arithmetic operator-product engines. A bootstrap lemma reduces its overdetermined bilinear commutativity systems to a distinguished odd charge D plus linear kernel problems – conditionally on even-spin vacancy, a machine-verified but finite-range input. For the Z_2 x Z_2 "paperclip" family we publish the spin-7 density P_8, prove [I_3,I_5]=[I_3,I_7]=[I_5,I_7]=0 identically in n, and verify the Lukyanov-Vitchev-Zamolodchikov four-screening characterization at spins 7 and 9 at sampled n. For the A_2^(2) family we give an exact Coulomb-gas dictionary to the dual Phi_{1,2}/Phi_{2,1} perturbations (that identification itself being classical, Zamolodchikov; Smirnov), whence the infinite hierarchy at all spins 6n+-1 follows by the Feigin-Frenkel affine-Toda theorem; our contribution is the certified dictionary, not a new all-spin construction. At c=1 the tower's defining charges are exactly quantum KdV at c_{1,4}=-25/2 – the all-odd sector of the Freeman-Hornfeck-West family: the hierarchy embeds at every odd spin, equality verified through spin 15, all-spin multiplicity one open. We exclude or disfavor the named non-KdV candidates at the tested loci and identify the self-dual degeneracy as L(1,0). Through spin 11, the generic centralizer of I_3 is, for all three Z_2 x O(N) "cylindrical" families, one-dimensional at odd and zero at even spins, with Q(t)-symbolic charges at spins 7, 9, 11 and an explained fiber jump at N=-2. We certify the published 2004 low-spin formulas and record five corrections (one presumed). All-spin statements are made only where proved, finite-range otherwise. We close with a corank-character conjecture linking the rank deficiencies of the bilinear systems to characters of screening kernels.
Read the full abstract