Machine Learning
iADD: Improving Alignment and Diversity in Diffusion Policy Optimization
Ashok Prasad Neupane, Saugat Adhikari, Pramish Paudel et al. (+2 more)
Reinforcement learning based post training of
diffusion models, such as Denoising
Diffusion Policy Optimization (DDPO), optimizes a reverse
diffusion process under a reward function. However, current approaches to reward optimizations do so at the cost of diversity and quality. In this paper, we provide better tradeoffs through careful theoretical considerations and method design. We analyze the theoretical framework and mathematically demonstrate that only-latter timestep updates of
diffusion model may be harmful for diversity contrary to the conclusions presented in a previous work. Additionally, we propose an incremental Feynman-Kac training based on strong theoretical foundations in order to achieve the best-yet
alignment-diversity tradeoffs. We perform extensive experiments and compare our method against related
diffusion policy optimization approaches in three different tasks and also provide strong ablations for each component, thus validating strong performance gains in both
alignment and diversity.
Read the full abstract ▾
All authors Ashok Prasad Neupane, Saugat Adhikari, Pramish Paudel, Ajad Chhatkuli, Danda Pani Paudel
Machine Learning
SAGE: Similarity-Based Cleaning of Poisoned Training Data from Verified Examples
Chaeeun Han, Soodeh Atefi, Yevgeniy Vorobeychik et al. (+1 more)
As machine learning increasingly relies on public, untrusted data sources, data poisoning attacks, which inject malicious examples into training data to induce misclassification of a chosen target, pose a growing threat. Existing defenses either assume zero ground-truth information about which examples are poisoned, or they assume access to a large set of examples verified to be clean. Satisfying the latter assumption incurs significant cost since reliable verification can be very resource- or labor-intensive. This cost is particularly high for clean-label attacks, where poisoned examples are visually indistinguishable from clean data. Since requiring a large set of verified examples is impractical, we propose relying on a small set of verified examples including both clean and poisoned ones, i.e., each example verified either to be clean or poisoned through inspection by a forensic expert. The challenge is then to detect poisons based on a set of verified examples that is so small that most classification models would overfit. To address this challenge, we propose Similarity-based Approach for Ground-truth-driven Exclusion (SAGE), which trains a generic feature extractor on a separate
dataset and then flags poisoned training examples using a non-parametric, similarity-weighted prediction based on the verified set. On standard
benchmarks against seven clean-label attack methods, we demonstrate that having access to even a handful of verified poisoned examples provides a substantial advantage. We also find that the distribution of verified clean examples across classes matters more than the number of verified examples.
Read the full abstract ▾
All authors Chaeeun Han, Soodeh Atefi, Yevgeniy Vorobeychik, Aron Laszka
Artificial Intelligence
Not All Experience Belongs in the Weights: Component Routing for Self-Improving GUI Agents
Beining Wu, Zihao Ding and Jun Huang
Self-improving GUI
agents keep the trajectories they produce and return them to the
agent, by fine-tuning or by
retrieval into the
prompt, and studies that compare the two destinations disagree. We attribute this to the unit of experience: a trajectory bundles items with different properties, so a conclusion about the bundle depends on its mix. To address this, (i) we introduce component routing, which splits the experience into locators, procedures, state facts and lessons and sends each component to the context or to the weights, compared on the same items across three backbone families, two environments and three seeds. One pool has two destinations: locators and lessons win in the weights, procedures and state facts in the context. (ii) We fit a rule in two properties measured before any training, recurrence and state-conditionality; it recovers the destination of a held-out backbone family in 24 of 24
cells, two interventions move a component toward the boundary, and routing by the rule beats every whole-trajectory baseline and, by +3.5 points on average, the better single destination of each backbone. (iii) We identify how training and producer-consumer differences change the value of the two destinations: note readout decreases after the same component is written into the weights, most for the items that recur most, context gains increase with the information gap, and weights gains decrease with the policy gap. Code and data will be released.
Read the full abstract ▾
Machine Learning
Inferring Multi-Timescale Neural Dynamics with Switching Linear Dynamical Systems
Lulu Gong, Yongxu Zhang and Shreya Saxena
Neural activity often exhibits multiple timescales that can vary with behavioral states and task conditions. Identifying these timescales from neural recordings is important for better understanding neural computation and function. However, traditional approaches based on autocorrelation fitting are difficult to scale to high-dimensional population recordings and can become unreliable when neural dynamics change with behavior. State-space models have been a powerful framework for modeling high-dimensional neural population activity through latent dynamical systems, but standard formulations and inference methods do not explicitly account for multiple timescales and therefore do not guarantee accurate recovery of the underlying temporal structure. Motivated by these questions, we introduce the Multi-Timescale Switching Linear Dynamical System (MTS-SLDS), a framework for identifying regime-specific latent timescales from continuous or spiking neural observations. MTS-SLDS combines a multi-lag moment initialization, which captures temporal structure across multiple observation lags, with regime-conditioned Laplace-EM inference, which reduces mixing of dynamical statistics across uncertain regimes. Characteristic timescales can then be extracted directly from the
eigenvalues of the learned latent transition
matrices. In synthetic and neural experiments with Gaussian and Poisson spike observations, MTS-SLDS accurately recovers timescales and switching structure over multiple
datasets.
Read the full abstract ▾
Computer Vision
Gueter Josmy Faure, Hao Ping Wang, Min-Hung Chen et al. (+1 more)
Processing long
videos with
Vision-
Language Models (VLMs) is bottlenecked by the quadratic cost of visual
tokens, making long-form inference prohibitively expensive. While single-axis compression methods mitigate this, they hit a hard efficiency floor because they treat spatial and temporal redundancy independently. We present VETO (
Video Efficient
Token Optimization for
Vision-
Language Models), a training-optional plug-in that eliminates this bottleneck through dual-axis compression: (i) an intra-frame compressor that merges semantically similar
tokens within each frame via optimal-transport inspired matching, and (ii) an inter-frame compressor that identifies and merges temporally redundant frames. The key design insight is hierarchical ordering: by first compressing spatial dimensions, VETO drastically reduces the cost of subsequent global temporal matching, bypassing the efficiency wall of single-axis approaches, with an advantage that grows with modern fully-fused
attention infrastructure. Empirically, VETO achieves up to 45% faster inference (e.g., on LLaVA-OneVision-7B) while preserving or improving accuracy. Under extreme
token starvation (10% budget), VETO outperforms VFlowOpt (54.9%), VisionZip (52.6%), and FastV (47.9%) with 55.7% accuracy. We demonstrate universal applicability across LLaVA-OneVision, InternVL-2.5, and LongVA, with zero-shot accuracy preserved or improved in all cases.
Read the full abstract ▾
All authors Gueter Josmy Faure, Hao Ping Wang, Min-Hung Chen, Winston H. Hsu
Distributed Computing
ePACT: Energy-Performance-Aware Commitment Tracking for LLM Serving
You Peng, Youhe Jiang, Chen Wang et al. (+1 more)
Reducing LLM serving energy does not by itself guarantee lower deployment cost when electricity procurement exposes operators to unfavorable deviations from preset commitments. We study hourly commitments with positive, potentially asymmetric costs for overuse and underuse, and formulate energy-Performance-Aware Commitment Tracking: minimize deviation costs subject to request-level service requirements. We implement ePACT, a two-level controller that adjusts serving capacity and GPU clocks as requests arrive. A global planner updates interval energy targets from measured consumption and the remaining hourly commitment. A local decision maker predicts candidate configurations' energy and completion times, checks predicted deadline misses, and selects among admitted configurations by asymmetric target-deviation cost, with a service-first fallback. Coarse-to-fine action search runs asynchronously with serving. We evaluate ePACT through single-hour comparisons, controller ablations, and full-day trace simulations for H20 and H200 GPU pools. In the 24-hour simulations, ePACT reduces the asymmetric deviation cost by $73.8\%$ and $75.7\%$ relative to vLLM while retaining near-vLLM SLO attainment. Mean absolute hourly deviations are $2.16\%$ and $2.31\%$, respectively.
Read the full abstract ▾
All authors You Peng, Youhe Jiang, Chen Wang, Binhang Yuan
Algebraic Geometry
Dihedral reflections and an infinite series of irrational Seshadri constants
Grzegorz Malara, Łukasz Merta, Justyna Szpond et al. (+1 more)
Laface and Ugaglia recently constructed an irrational one-point Seshadri constant on the blow-up of $\mathbb{P}^2$ at nine very general points by combining a dihedral orbit on $\mathbb{P}^1\times\mathbb{P}^1$, a sequence of de Jonquières transformations, and a reflection argument along a $(-4)$-curve with balanced normal bundle. We show that the same mechanism extends uniformly to every odd integer $n\geq 5$. For $n=2k+1$ we prove $$ \varepsilon\bigl(\mathcal{O}_{\mathbb{P}^1\times\mathbb{P}^1}(n-4,1);p_1,\ldots,p_{2n}\bigr)=\sqrt{\frac{n-4}{n}} $$ at $2n$ very general points, and already at a very general free orbit of a fixed dihedral group of order $2n$. For every odd $n\geq 7$ this produces an explicit ample line bundle on the blow-up of $\mathbb{P}^2$ at $k+7=(n+13)/2$ very general points whose one-point Seshadri constant at a very general point equals $$ 2\sqrt{n(n-4)}. $$ More precisely, after one quadratic transformation we obtain the ample divisor $$ L_n=(3n-4)H-nE_1-(n-2)(E_2+\cdots+E_5)-4(E_6+\cdots+E_{k+5})-2(E_{k+6}+E_{k+7}), $$ with $L_n^2=4n(n-4)$ and $\varepsilon(L_n;x)=\sqrt{L_n^2}$. We also isolate an abstract balanced-reflection principle underlying the construction: a nef class on a special fiber can be reflected across a rational curve of square $-2a$ whenever the curve has normal bundle $\mathcal{O}_{\mathbb{P}^1}(-a)^{\oplus 2}$ in the total space, and the reflected class is nef on very general fibers.
Read the full abstract ▾
All authors Grzegorz Malara, Łukasz Merta, Justyna Szpond, Marcin Zieliński
Quantum Physics
Non-Markovian effects on informational steady states
Jacob Werner
Informational steady states (ISSs) arise when the rate at which information is gained from continuous measurements is exactly balanced by the rate at which information from earlier measurements is lost to the environment. While ISSs have been studied in Markovian settings, relatively little is known about how
memory effects influence their formation and steady-state properties. In this work, we investigate non-Markovian effects on ISSs within the framework of continuously monitored collision models. We model the environment as a sequence of two-
qubit ancillas and find that, for this specific model, the steady-state information gain per measurement is negatively correlated with the degree of non-Markovianity. Numerical results further characterize how the system-ancilla and ancilla-ancilla interaction parameters influence both the degree of non-Markovianity and the steady-state information gain.
Read the full abstract ▾
Artificial Intelligence
Q-Learning for Reachability in MEC-Free MDPs
Lu-Chin Chang and Suguman Bansal
Reinforcement learning (RL) for reachability specifications is fundamental to sequential decision-making. Prior work establishes asymptotic convergence to optimal policies, but only through model-based methods that must explicitly estimate the transition probabilities of the underlying Markov Decision Process (MDP). We present
Quasar, the first model-free
algorithm with asymptotic guarantees for reachability on the fragment of MDPs free of non-terminal maximal end components (MECs), a building block to which every MDP reduces by the standard MEC quotient. Our
algorithm follows the classical Q-learning approach, using temporal-difference updates to converge to an optimal policy without ever learning the transition probabilities. The resulting learner reduces the
memory footprint from the O(|S|^2|A|) that model-based methods require to O(|S||A|). On the standardized Quantitative Verification
Benchmark Set, our
algorithm converges to the optimal policy with orders of magnitude fewer samples than the previous model-based state-of-the-art. Together these results are a concrete step toward the practical deployment of reachability learning and, with it, of specification-guided RL.
Read the full abstract ▾
Artificial Intelligence
RealCompanion: Benchmarking Human Understanding from Reasoning over Longitudinal Real-World Conversations
Arman Behnam, Sunglyoung Kim, Jiayi Yu et al. (+2 more)
An AI companion that talks with someone for months should come to understand them. It should know who they are, remember what they said, and recognize when something from the past matters now. Testing this needs real conversations, but real conversations are private, so existing
benchmarks use invented people and invented questions. We release RealCompanion, ten real relationships between people and an AI companion, with 27,218 messages over up to 120 days. For each person, we release the full conversation, a profile, a persona, chat test items, and question test items. Every label points to the messages that support it, and every chat label comes with the
reasoning that produced it. The real data shows three things. First, people rarely refer back. Only 3.4% of their messages depend on something said earlier, and when one does, the earlier message is usually far away (a median of 2,157 messages back). Averages hide this. Looking at the most recent messages finds the needed one 95.9% of the time overall, but only 2.2% of the time when it is far back. Second, AI systems cannot tell when the past matters. The detectors we tested barely beat chance on real messages, and when the same earlier messages are labeled "
memories" instead of "earlier messages", models bring up the past 10 to 14 percentage points more often, even when nothing from the past is needed. Third, AI systems read more into a person than the person revealed. Three
agent systems rebuild each persona equally well (F1 0.71). They see the person, and then imagine more. Understanding a person depends on knowing when their past matters and where what they shared ends, and only real conversations can test it.
Read the full abstract ▾
All authors Arman Behnam, Sunglyoung Kim, Jiayi Yu, Eric Huang, Liangwei Yang
Analysis of PDEs
Inverse Energy Cascade and Non-uniqueness for the Advection-Diffusion Equation
Alexey Cheskidov and Xiaoran Liu
We prove non-uniqueness for the advection-
diffusion equation with a drift in a logarithmically near-critical class below $L_t^2L_x^\infty$. The first construction is a
diffusion-assisted inverse cascade of scalar energy that diverges as $t \to 0$. The second uses compressed inverse mixing, with
diffusion acting perturbatively, and produces a bounded
parabolic solution with the energy jump at the initial time. In both constructions the drift and scalar are smooth for positive time. We also consider the drift in $L^{2,\infty}_tL^\infty_x$ and solutions satisfying the energy inequality starting from almost every time. In this setting, we prove uniqueness of solutions with bounded energy, and construct non-unique solution with energy blowing up as time goes to zero.
Read the full abstract ▾
Computer Vision
GIFTBench: Diagnosing Generalization in Image Forgery Localization and Informing Model Design
Baoke Dou, Ziye Wang, Hao Wang et al. (+5 more)
Reliable evaluation of image forgery localization (IFL) requires assessing models under diverse distribution changes, yet existing
benchmarks often cover limited manipulation conditions or entangle multiple factors in cross-
dataset evaluation. Consequently, aggregate performance provides an incomplete view of localization generalization. We introduce GIFTBench, a multi-axis
benchmark of 115,013 manipulated images with pixel-level annotations spanning manipulation source, semantic target, editing operation, and composition complexity. GIFTBench supports axis-specific transfer analysis and evaluation on twelve external
datasets. Its diagnostic studies reveal asymmetric cross-source transfer, recall-dominated failures, and heterogeneous degradation across semantic, operational, and compositional changes. Beyond diagnosis, the scale and diversity of GIFTBench provide a substantially broader training distribution than conventional IFL
datasets. Training representative localizers on GIFTBench consistently improves their aggregate transfer to external
datasets, showing that the
benchmark serves not only as an evaluation tool but also as an effective training resource for cross-domain localization. Guided by the diagnostic findings, we further develop ForenScope, a detection and localization framework combining classification-adapted representations with multi-depth, multi-scale spatial features, learned layer fusion, and selective coarse-scale conditioning. Experiments show improved cross-
dataset localization while retaining image-level detection capability. The GIFTBench
dataset showcase page is available at https://giftbench-preview.doudoudouya337.chatgpt.site.
Read the full abstract ▾
All authors Baoke Dou, Ziye Wang, Hao Wang, Guoqing Cai, Wende Tan, Chenyang Si, Liucheng Guo, Yueming Lyu
Quantum Physics
QUFIG: GNN-Based Prediction of Quantum Fault Injection Vulnerabilities with Gate-Level Precision
Shihan Zhao, Qiying Li, Ben Dong et al. (+2 more)
The growing scale and accessibility of
quantum hardware exposed new reliability and security challenges in the
quantum computing workflow, such as the run-time fault injection attacks in cloud-based
quantum computing platforms. However, existing works fail to identify vulnerabilities with gate-level precision or adapt to run-time environments. In this work, we formulate gate-level fault analysis as a learning-guided prioritization problem under restricted fidelity budgets. The framework uses a
circuit-DAG-based GNN backbone to predict the vulnerability score of each gate to each type of injected fault, defined as the impact of the gate-fault pair on
circuit fidelity. The gate-fault pairs are then ranked by their vulnerability score. Experiments on QASMbench and HamLib MaxCut show that QUFIG recovers high-impact vulnerable gate-fault pairs with fewer inspections than random and depth-based heuristics. Our results show that QUFIG can reduce the number of gates requiring inspection by 2.9–19.8% while maintaining effective fault identification, allowing
quantum circuit designers to identify vulnerabilities and apply targeted defenses more efficiently.
Read the full abstract ▾
All authors Shihan Zhao, Qiying Li, Ben Dong, Qian Wang, Yuntao Liu
Geometric Topology
Sections, pseudosections, and multisections of Lefschetz fibrations
R. Inanc Baykur and Noriyuki Hamada
We show that not every
symplectic Lefschetz fibration over the $2$–sphere admits a smooth section, settling a long-standing open problem. To prove this, we develop a method for studying sections via lifts of monodromy factorizations with point-pushing maps and the induced handle decomposition of the total space, and apply it to an infinite family of Lefschetz fibrations. In contrast, we show that every Lefschetz fibration over $S^2$ admits positive multisections. Some of these examples admit no multisections with spherical components. We also construct signature-zero Lefschetz fibrations of every odd genus $g\geq 7$, which in turn give
symplectic Lefschetz fibrations of any prescribed signature in each of these genera.
Read the full abstract ▾
High Energy Physics (Theory)
From Hidden Zeros to Exact Splittings and New Zeros
Yang Li and Laurentiu Rodina
Hidden zeros impose constraints on scattering amplitudes beyond their familiar factorization properties. We explore their structure in two related directions. We first promote hidden zeros to explicit splitting formulas that organize full amplitudes at generic kinematics. Ordered $\mathrm{Tr}(\phi^3)$ amplitudes, nonlinear sigma model (NLSM) amplitudes, and cosmological wavefunctions share the same defining exact splitting structure $\mathcal A=\sum_\alpha R_\alpha\, {\mathcal J}_{\alpha,L}{\mathcal J}_{\alpha,R} {\mathcal J}_{\alpha,\mathrm{out}}$, adapted to any hidden zero. Each prefactor $R_\alpha$ is a condition of the hidden zero locus, so this decomposition manifests the hidden vanishing term by term. The remaining factors are lower point amplitudes or currents. Using these representations, we can systematically find new zero loci beyond the usual ones. In particular, we obtain five new all-multiplicity zero families. We also find continuous families of zeros: the coefficients of their defining equations can vary continuously while the amplitude remains zero.
Read the full abstract ▾
Quantum Algebra
Ordinary modules for affine vertex operator superalgebras
Huaimin Li and Qing Wang
Let $\mathfrak{g}$ be a basic classical Lie superalgebra and let $\widehat{\mathfrak{g}}$ be the corresponding affine Lie superalgebra. In this paper, we first prove that a Cartan subalgebra acts semisimply on ordinary modules for the simple affine vertex operator superalgebra $L_{\widehat{\mathfrak{g}}}(k,0)$ at boundary admissible level $k$. Then we prove that the category $\mathcal{O}_{k}^{ord}(\mathfrak{g})$ of ordinary $L_{\widehat{\mathfrak{g}}}(k,0)$-modules is finite, semisimple and $\mathcal{O}_{k}^{ord}(\mathfrak{g})$ is exactly the category $KL_k(\mathfrak{g})$ of finite-length generalized modules for the affine vertex operator superalgebra $L_{\widehat{\mathfrak{g}}}(k,0)$.Thus $\mathcal{O}_{k}^{ord}(\mathfrak{g})$ is a braided tensor supercategory. Furthermore, we obtain the
rigidity of the supercategory $\mathcal{O}_{k}^{ord}(\mathfrak{g})$ and thus it is a ribbon supercategory. Finally, we conclude that $\mathcal{O}_{k}^{ord}(\mathfrak{g})$ is a ribbon fusion supercategory.
Read the full abstract ▾
Computational Engineering
CODesign: Consistency from Data to Trajectory in All-Atom Protein Binder Co-Design
Yuanle Mo, Bo Qiang, Haitao Lin et al. (+4 more)
The central challenge in de novo
protein design is generating plausible, mutually compatible structures and sequences, such that each designed sequence folds into its intended structure and the structure accommodates that sequence. Compared to typical two-stage design methods, which decouple the modeling of the interdependent modalities, co-design models improve the cross-modal consistency by jointly generating sequences and structures. However, naively generating sequences and structures simultaneously does not ensure their consistency. To address this challenge, we propose CODesign framework. We improve data consistency by generating approximately 105,000 consistency-distilled dimers. We further promote consistency through a
multimodal joint flow model that captures the joint distribution of sequences, backbone structures, and local atomic configurations, together with a consistency-aware joint resampling strategy that iteratively refines sequences and side chains. Experiments show that CODesign achieves state-of-the-art performance with the highest in silico success rates on both
protein- and ligand-target binder design. Ablation studies also demonstrate our distilled
dataset increases performance by 70.9%, which can be further improved by our proposed resampling mechanism with negligible additional computational cost. Code, model weights and the new
dataset will be completely open-source.
Read the full abstract ▾
All authors Yuanle Mo, Bo Qiang, Haitao Lin, Qinghan Wang, Gang Du, Odin Zhang, Pheng Ann Heng
Quantum Physics
Efficient quantum phase estimation with adaptive entanglement-assisted Hadamard test
Hengzhun Chen, Benchi Zhao and Yingzhou Li
The
entanglement-assisted Hadamard test (EHT) is a practical method for estimating a
quantum phase by amplifying the phase signal. However, the feasible amplification is fundamentally limited by the accuracy of the reference phase, such that the method is inefficient in the high-precision regime. In this work, we propose an
algorithm, called adaptive
entanglement-assisted Hadamard test (AEHT), that iteratively refines the reference phase, enabling progressively stronger amplification as the iteration goes by. We further consider the imperfect eigenstate preparation scenario, where a systematic bias is unavoidable when estimating the
quantum phase with the conventional EHT. Such a bias can be suppressed by the proposed AEHT. Moreover, taking physical implementation into consideration, we adopt device-restart count to measure the cost of
quantum phase estimation, rather than shot count. The numerical experiments confirm the effectiveness of the proposed AEHT compared with conventional methods under this measure. By unlocking the full amplification power of large entangled states, this work offers an efficient method to estimate high-precision
quantum phase on near-term
quantum processors.
Read the full abstract ▾
Chemical Physics
Nonadiabatic Quantum Dynamics of Hexatriene via the Extended Hubbard-Peierls Model and Time-Dependent DMRG
Timothy N. Georges, Darren J. Valentine and William Barford
Propagating the electronic and
nuclear degrees of freedom of a molecule simultaneously is a challenging task, because the number of vibronic states grows exponentially with system size. We use the adaptive time-dependent density
matrix renormalization group to perform simulations of the extended Hubbard-Peierls
Hamiltonian, in which both the electronic and
nuclear degrees of freedom are quantized. We describe our approach, and demonstrate it through converged calculations of the nonadiabatic dynamics of hexatriene. This simulation predicts a lifetime of the `bright' $\mathrm{1B_u}$ electronic state of approximately 40 fs. Our predictions for the population of the $\mathrm{1B_u}$ electronic state are in qualitative agreement with those obtained using the linear vibronic coupling model derived from the extended Hubbard-Peierls
Hamiltonian in previous work.
Read the full abstract ▾
Computational Engineering
A pyramidal ISRU lunar habitat design based on topological interlocking of sintered regolith blocks
Lukas Schnelle, Kai-Uwe Schröder, Alice C. Niemeyer et al. (+1 more)
In this framing study, we propose a new lunar-habitat design based on topological interlocking of the building blocks. The design combines block geometries that permit interlocking without binders or connectors with a pyramidal overall structure. This approach is well suited for ISRU-enabled extraterrestrial construction. Finite element simulations employing the Drucker-Prager model confirm that habitats manufactured from lunar regolith using this concept are tolerant to local failures or missing blocks. We further suggest that this approach can be extended and improved by optimising the geometry of the interlockable building blocks.
Read the full abstract ▾
All authors Lukas Schnelle, Kai-Uwe Schröder, Alice C. Niemeyer, Yuri Estrin
Software Engineering
CONTRA: Discovering and Qualifying Behavior-Changing Questions for Selective Clarification in LLM Code Generation
Zheng Fang, Yongmin Li, Yichang Zhang et al. (+5 more)
Coding
agents can generate code that appears correct but implements behavior the user never intended. This mismatch can arise when an
agent silently resolves underspecified requirements through its own assumptions. As subsequent development builds on these assumptions, correcting the resulting behavior can become increasingly costly. Early clarification can help prevent such mismatches, but unnecessary questions can interrupt developers and slow down development. Existing methods struggle to identify key clarification questions while avoiding unnecessary ones. Therefore, we propose CONTRA, a training-free method that combines broad question discovery with semantic and execution-based question qualification. CONTRA first generates candidate questions and filters out those unrelated to required behavior or already resolved by the requirement. For each remaining question, it generates programs conditioned on two plausible answers and checks for stable behavioral differences on shared inputs. It then uses the interaction history to select among qualified questions or stop asking. Experiments on ClarifyCodeBench show that CONTRA achieves the highest F1 with all four coding
agents, exceeding the best baseline macro-average F1 by 13.88 percentage points. With the same LLM and evaluation protocol, CONTRA also achieves higher clarification recall and F1 than the coding harnesses Claude Code and OpenHands. To support practical use, we also implement CONTRA as a Claude Code plugin that integrates selective clarification into everyday development.
Read the full abstract ▾
All authors Zheng Fang, Yongmin Li, Yichang Zhang, Dongming Jin, Haoyu Wang, Shuai Wang, Zhi Jin, Ge Li
Cryptography and Security
The Innocent Courier: Covert Exfiltration Through Legitimate LLM Web Fetching
Alessandro Pegoraro, Daryan Merx, Phillip Rieger et al. (+1 more)
With the increasing capabilities of Large-
Language-Models (LLMs) and LLM-based
agents, users are increasingly using them to solve everyday problems, such as answering e-mails or providing programming support. Existing work has extensively investigated security and
privacy risks, such as
prompt injections and the disclosure of sensitive data to chatbot providers. While various solutions were developed to address these risks, including input structuring to prevent
prompt injections or deploying local LLMs to avoid sharing confidential data with chatbot operators, LLMs also pose the risk of leaking confidential data to third parties. In this paper, we demonstrate with LLMLeak a novel attack vector where malicious software that runs locally but cannot communicate directly with the internet abuses LLMs to establish a covert channel. While inputs that instruct the LLM to send data directly via generated code are easy to detect and network libraries are typically restricted, LLMLeak relies only on the LLM's tool to fetch websites for further information. A malicious software component on the client side embeds a secret into a URL. It presents the referenced website as providing information required for a benign task, such as migrating a software library. When the LLM accesses the URL, the attacker receives the encoded secret through an attacker-controlled DNS or web server. We perform an extensive evaluation on eleven open-parameter models, observe an attack success rate of 79.7%, and also conduct a case study on real-world chatbots, demonstrating the relevance of LLMLeak.
Read the full abstract ▾
All authors Alessandro Pegoraro, Daryan Merx, Phillip Rieger, Ahmad-Reza Sadeghi
Computation and Language
A Matryoshka Hierarchical RAG for Efficient Multi-Hop Question Answering
Gianluca Bonifazi, Christopher Buratti, Michele Marchetti et al. (+5 more)
Retrieval-Augmented Generation (RAG) systems for multi-hop Question Answering (QA) must balance
retrieval quality with computational cost. This cost is incurred during indexing time, through the use of expensive Knowledge Graphs (KGs) or Large
Language Models (LLMs) to generate summaries, or during querying, through iterative LLM-driven
retrieval. To reduce it while maintaining
retrieval quality, we present MatRAG, a hierarchical framework that combines RAG systems with Matryoshka Representation Learning (MRL). MatRAG addresses both kinds of cost by aligning the semantic hierarchy of a clustering structure with the nested structure of MRL. Specifically, it organizes the corpus of documents into a Directed Acyclic Graph (DAG) of clusters with progressively coarser granularity. Each level is indexed by a lower Matryoshka dimension. MatRAG pairs an iterative, top-down traversal of the DAG with an entity-driven mechanism that controls the hop budget and re-ranks candidates. We evaluated MatRAG on three standard multi-hop QA
benchmarks against seven representative baselines. MatRAG outperforms its strongest competitors in terms of
retrieval quality; furthermore, it reduces indexing costs by avoiding KG construction and LLM-based summarization, and lowers query-time costs through dimension-aware similarity.
Read the full abstract ▾
All authors Gianluca Bonifazi, Christopher Buratti, Michele Marchetti, Federica Parlapiano, Giulia Quaglieri, Davide Traini, Domenico Ursino, Luca Virgili
Artificial Intelligence
VideoEvolve: Evolving Agent Harnesses for Video Temporal Grounding
Bingjun Luo, Yuhuan Fan, Jialin Guo et al. (+1 more)
Video temporal grounding aims to localize events in
videos from natural-
language queries. For
agents built around frozen
video-
language models, the harness determines how queries guide temporal predictions and how those predictions are refined. Manually refining these harnesses requires diagnosing grounding failures and coordinating changes to both
agent workflows and instructions. We introduce VideoEvolve, a framework that automatically evolves
agent harnesses for
video temporal grounding. VideoEvolve uses a Cloze-Structured Harness Representation that preserves stage interfaces while leaving
agent workflows and instructions open to
evolution. Branch-Guided Harness
Evolution preserves promising code branches for continued refinement, using execution feedback to guide local edits and validation to determine which improvements are carried forward. Experiments demonstrate improved grounding performance across multiple
benchmarks. Component analyses identify instruction refinement as a consistent source of gains, while the benefits of evolved code vary across evaluation settings. Together, these results support automated harness
evolution as an effective approach to improving
video temporal grounding. Code is available at https://github.com/bingjunluo/VideoEvolve .
Read the full abstract ▾
All authors Bingjun Luo, Yuhuan Fan, Jialin Guo, Siqi Li
Machine Learning
Physics-Refined Spatiotemporal Forecasting on Open-Boundary Hydrologic Graphs
Haoyang Jiang, Zhengui Wang, Shenghan Gao et al. (+3 more)
Spatiotemporal
forecasting on hydrologic graphs is especially prone to instability in open-boundary systems, where the forecast domain exchanges fluxes with an unobserved exterior. In such systems, boundary nodes receive external forcing, e.g., upstream inflows in rivers or
tidal signals in coastal regions, that is typically unavailable at prediction time. The absence of this information can compound errors as forecasts unfold in an autoregressive fashion, leading to inferior long-horizon performance. This paper dissects this instability issue by exploring two questions. 1) What boundary forcing enters the forecast domain when information beyond the boundary is missing? 2) How should this forcing propagate through the domain without incurring error amplification under autoregressive rollout? To address both, we propose a new computing framework comprising two key components. First, to compensate for the boundary forcing, our framework learns ghost node proxies from the boundary and interior nodes, striving to approximate unobserved external inputs. Second, to control error accumulation from these learned proxies, we leverage two physics refiners. In particular, one refiner enforces local consistency by aligning ghost proxies with their two-hop neighbors (i.e., boundary nodes and their immediate interiors). The other refiner enhances global stability by correcting the model forecasts through a physics-guided graph neural operator, reducing long-horizon numerical drift. Two real-world hydrologic graphs are employed for empirical evaluation. Comparative results show that our proposal enjoys higher prediction accuracy and long-horizon stability over both learning-based and physics-informed model competitors.
Read the full abstract ▾
All authors Haoyang Jiang, Zhengui Wang, Shenghan Gao, Y. Joseph Zhang, Xingquan Zhu, Yi He
Materials Science
Weak non-Landau-type contributions to the diamagnetism of graphite
A. V. Nikolaev, A. V. Bibikov, M. Ye. Zhuravlev et al. (+1 more)
The exceptionally large diamagnetic susceptibility of a single-
crystal graphite in the direction perpendicular to the
graphene layers is caused by the Landau levels. However, there are also other contributions to the
magnetic susceptibility. In particular, in metals the Pauli paramagnetism is leading, whereas in dielectrics the diamagnetic Langevin and paramagnetic Van Vleck terms are essential. These components of
magnetic susceptibility are also present in graphite. In this work, we calculate the magnitudes of all these small contributions from first principles. It is found that the Pauli paramagnetism is negligible (2.3 x 10^{-9} emu/g), whereas the other diamagnetic contribution - arising from the Langevin and Van Vleck mechanisms - is comparable to the in-plane experimental diamagnetism of graphite and to the diamagnetic response of C60 and C70} fullerenes. This diamagnetic contribution proves to be slightly anisotropic (-2.93 x 10^{-7} emu/g along the x- or y-axes and -2.24 x 10^{-7} emu/g along the $z-$axis), with the averaged value -2.7 x 10^{-7} emu/g.
Read the full abstract ▾
All authors A. V. Nikolaev, A. V. Bibikov, M. Ye. Zhuravlev, L. L. Tao
Artificial Intelligence
TopK-Guided: Adaptive, Budget-Aware Activation Sparsity for Efficient LLM Inference
Mukund Agarwalla and Chih-Jen Lin
Activation sparsity speeds up large
language model (LLM) inference by setting unimportant activations to zero so that the corresponding computations can be skipped. Existing training-free methods, however, make different trade-offs: threshold-based methods such as TEAL adapt the sparsity level to each
token but do not tightly control the realised sparsity, while TopK-based methods such as WINA enforce a fixed sparsity level but use the same sparsity budget for every
token. Both also apply the same budget across
transformer blocks, despite large differences in block sensitivity. We introduce TopK-Guided, a training-free method that addresses both limitations by combining bounded
token-level sparsity adaptation with sensitivity-aware block-level budget allocation. Across Llama-2 and Llama-3 models, TopK-Guided consistently improves perplexity and downstream accuracy over TEAL and WINA while preserving essentially the same sparsitydependent projection compute as WINA, with the largest gains at high sparsity. Ablations show that both components provide complementary improvements.
Read the full abstract ▾
Computer Vision
OneStreamer: Unifying Perception, Memory, and Proactive Response in Streaming Video Interaction
Xiangyu Zeng, Yuandong Yang, Zhiqiu Zhang et al. (+21 more)
Streaming
video LLMs must retain evidence before its relevance to future tasks is known and respond when sufficient evidence becomes available. The challenge is to form reusable factual
memory without compromising real-time perception. We introduce OneStreamer, which jointly learns query-independent evidence recording and task response through a shared proactive generation process. Its Proactive Hierarchical Caption
Memory (PHCM) produces time-grounded local-detail captions and summaries of completed events. Streaming caption targets supervise the interpretation of observed
video prefixes during training. At inference, model-generated records complement a recent visual window, providing reusable factual context without revisiting historical visual features. Proactive State Transition Learning (PSTL) reduces the dominance of repeated waiting states by preserving supervision at all output anchors and selecting representative state-change and state-persistence
tokens. We further develop a streaming data synthesis pipeline that aligns output content and timing with available evidence. Combining the resulting streaming captions and QA with cleaned open-source data yields OneStreamer-1M, a broad-coverage streaming
video interaction
dataset with over one million records spanning diverse tasks. Our 4B model achieves the best results among the compared methods across all eight evaluated streaming
video understanding
benchmarks. Ablations show that retaining generated captions improves historical QA without degrading real-time perception. PSTL also outperforms dense state supervision while supervising only 27.5% of annotated state
tokens. Together, these results support proactive generation as a shared learning interface connecting perception,
memory formation, and timely response in streaming
video interaction.
Read the full abstract ▾
All authors Xiangyu Zeng, Yuandong Yang, Zhiqiu Zhang, Yuhan Zhu, Xinhao Li, Qingyi Si, Dingyu Yao, Changlian Ma, Haoran Chen, Xinyu Chen, Yansong Shi, Junhao Zhou, Yifei Li, Jun Zhang, Chuanyu Qin, Chenxu Yang, Xinlei Yu, Kun Ouyang, Yuchen Shao, Qianshan Wei, Changhai Zhou, Jun Gao, Jiaqi Wang, Limin Wang
High Energy Physics (Experiment)
A-GHOST: High-rate streaming of trigger-level data to programmable GPU inference
I. Xiotidis, N. Clarke Hall, M. S. Larson et al. (+11 more)
A-GHOST (A Global Heterogeneous Online Scouting Trigger) is an R&D effort investigating high-rate streaming readout architectures for high-energy physics (HEP) experiments. The central idea is to convert the deterministic front-end and fixed-latency processing close to the detector from a decision-making platform to an aggregation and streaming source. Compact trigger-level data are streamed to a GPU-enabled backend, where substantially more complex
algorithms can run beyond the resource and latency envelope of the hardware trigger. This paper presents a proof-of-concept backend using the NVIDIA IGX Thor development kit. A software rate-controlled transmitter sends data through a QSFP interface and external loopback cable to a second QSFP interface, allowing the network-to-GPU path to be studied before integration with FPGA sources. NVIDIA DAQIRI enables reception directly into GPU-accessible
memory, while a custom CUDA kernel reassembles packet payloads into persistent, contiguous TensorRT input windows without data-type conversion, which is handled by the models. Using a HEP-derived event representation consisting of the ten leading calorimeter clusters, the backend scales from 40 to 100 Gbps while sustaining a constant input rate and an inference latency of 0.118 ms at the p95 interval. The system operates for multiple hours without drops or inference errors. Generative and temporally aware neural networks demonstrate workloads beyond those normally deployed in fixed-latency FPGA trigger systems. The result establishes A-GHOST as a basis for future FPGA-to-GPU streaming readout systems.
Read the full abstract ▾
All authors I. Xiotidis, N. Clarke Hall, M. S. Larson, R. Gurunathan, C. Burdick, T. Du, N. Konstantinidis, K. Kordas, D. Leshchev, D. W. Miller, V. A. Petrovic, D. Sampsonidou, A. Thompson, T. Wengler
High Energy Physics (Theory)
Tree-Level Interaction Vertices in the Linear Sigma Model with Quarks under Global Rotation
Luis A. Hernández, Eduardo Lazcano and R. Zamora
We derive the tree-level interaction vertices of the linear sigma model with
quarks in the presence of global rotation. Starting from the scalar and fermionic field solutions in a rotating background, we construct the corresponding spectral representations and use them to evaluate the mesonic self-interactions and Yukawa couplings of the model. We show that rotation modifies the usual momentum-space structure of the interaction vertices: conservation of longitudinal momentum is accompanied by angular-momentum selection rules, while the transverse dynamics is encoded in rotation-induced radial form factors involving products of Bessel functions. The resulting expressions provide a set of interaction rules adapted to the
symmetries and finite geometry of the rotating system. Although derived within the linear sigma model with
quarks, the structures obtained for scalar self-interactions and scalar-
fermion couplings can be extended to a broader class of
quantum field theories under global rotation. These results provide the building blocks for perturbative calculations and contribute toward a microscopic
quantum-field-theoretical description of interacting matter in rotating environments.
Read the full abstract ▾