arXiv:2512.18368v2 Announce Type: replace Abstract: Scaling imitation learning to diverse multi-task robot manipulation remains challenging due to suboptimal demonstrations, behavioral multi-modality, and destructive interference across tasks. While skill-based methods offer a promising direction by decomposing behaviors into reusable abstractions, existing approaches often learn skills that are either biased toward linguistic structure or lack semantic alignment across tasks, limiting generalization. In this work, we propose AtomSkill, a novel framework that learns a semantically aligned Atomic Skill Space from demonstrations and enables robust long-horizon execution through keypose imagination. Our method introduces: (1) semantic contrastive skill alignment, which partitions demonstrations into variable-length atomic skills and employs a contrastive objective to jointly enforce semantic consistency and temporal coherence, yielding a compact and reusable skill library; and (2) action decoding with keypose imagining, where the policy predicts both a skill's terminal keypose and immediate actions, thereby supporting progress-aware skill transitions. During inference, an atomic skill diffusion sampler generates plausible skill sequences, while predicted keyposes autonomously trigger smooth skill chaining. Extensive experiments in simulation and real-world settings show that AtomSkill consistently outperforms state-of-the-art imitation learning and skill-based baselines. Project page: https://atom-skill.github.io.
Science Journals
arXiv:2602.16403v2 Announce Type: replace Abstract: Traditional models of slow slip events (SSEs) oversimplify fault geometry, although imaging shows subduction faults are segmented and complex. We examine how fault interactions control slip behavior using 3-D quasi-dynamic earthquake sequence simulations of two parallel faults with uniform rate-weakening friction accelerated by hierarchical matrices. Four regimes emerge-periodic earthquakes, coexisting SSEs and earthquakes, only SSEs, and complex sequences-whereas with the same friction condition a single planar fault produces only earthquakes. We quantify interaction using the maximum Coulomb stress induced on a target fault by a spatially uniform unit stress drop on a neighboring fault. Because the stress drop is normalized, the metric depends only on geometry and is independent of friction, allowing extension to arbitrary fault systems. SSEs occur only at intermediate fault interaction strengths. At low interaction strengths, the system produces regular, periodic earthquakes. At high interaction strengths, fault interactions generate complex earthquake sequences with irregular recurrence and variable magnitudes. Simulations reproduce observed moment-duration scaling and show sensitivity to detection thresholds. These results demonstrate geometric complexity alone generates both slow and fast earthquakes through evolving traction heterogeneity.
arXiv:2607.01562v1 Announce Type: new Abstract: Classical software effort distribution models, including the PNR family and Parr alter native curve, were designed to describe the time distribution of development effort under an implied staffing pattern. Their direct use in agile environments is limited when team capacity is fixed, partially fixed, or externally constrained, the original curve may prescribe a staff demand that the organization cannot allocate. This paper proposes a compact refactoring of Parr model as a capacity-aware forecasting layer for agile projects. The contribution is deliberately narrower than a full causal theory of project dynamics. A normalized Parr shaped latent effort demand is combined with an observed or planned capacity trajectory. The resulting model forecasts aggregate progress, completion time, capacity deficit, and capacity slack without assuming that the same internal activity path is followed under resource restriction. The model uses a small parameter set such as total effort K, a Parr shape parameter, an origin constant c that can match nonzero initial staffing, and the capacity trajectory. A discrete sprint formulation is provided, together with a calibration method from ordinary Scrum records and a rolling origin validation protocol against simple management baselines.
arXiv:2607.01571v1 Announce Type: new Abstract: Chain-of-thought (CoT) reasoning enables large language models (LLMs) to solve complex problems by generating intermediate reasoning steps. While much attention has been paid to the length and content of these reasoning chains, far less is known about their internal geometry. We study the \emph{geometry} of CoT trajectories in the hidden state space of transformer models, formalizing each reasoning chain as a discrete curve in $\mathbb{R}^d$ and characterizing it through spectral, positional, and kinematic geometric functionals. We introduce the effective dimension $d_\rho$ as a measure of trajectory complexity and show theoretically that trajectories with flatter eigenvalue spectra correspond to harder tasks, as they explore more of the hidden dimensions. Lastly, we explore how kinematic features of the trajectory, mean position, positional dispersion, initial and current hidden states, mean velocity, mean speed, and speed dispersion, can be used to predict solution correctness before generation is complete, and may inform future early-stopping strategies. Experimentally, on mathematical reasoning problems from the MATH500 dataset, $d_\rho$ achieves $0.93$ AUC in distinguishing easy from hard problems, while kinematic features potentially can predict correctness from only the first $20\%$ of generated tokens. These correctness signatures transfer across questions of varying difficulty, establishing that the shape of a model's internal reasoning trajectory is a principled window into both task hardness and solution quality.
arXiv:2607.01785v1 Announce Type: new Abstract: Next activity prediction helps service-oriented processes anticipate upcoming steps before delays, exceptions, or service-level risks occur. Most existing methods assume classical single-case event logs, whereas real service processes often involve events shared by multiple typed business objects. Object-centric event logs (OCELs) capture such interactions, but current predictors remain limited. Flattening-based approaches lose cross-object context, and native OCEL graph-based approaches encode multi-object events through pairwise relations. Existing models also do not jointly capture event-driven object state changes, inter-event timing, and global execution patterns. We propose EHHN, an Event-driven Heterogeneous Hypergraph Network for object-centric next activity prediction. EHHN represents each prediction prefix as a heterogeneous hypergraph, where event--object hyperedges bind retained co-participating objects and a lifecycle hyperedge groups the primary object's observed lifecycle events. Based on this representation, EHHN uses a dual-stream architecture in which a micro-spatial stream models event-driven object-state evolution and a macro-evolution stream captures temporal dynamics using retrieved global prototypes. The two streams are fused to predict the next activity. Experiments on four public OCEL benchmarks against nine baselines show that EHHN achieves the best accuracy and macro F1-score on all datasets, with improvements of up to 8.1 and 12.4 percentage points over the strongest baselines. Compared with the strongest OCEL-native graph baseline, EHHN also reduces peak GPU memory by up to 24 times. Code is available at https://github.com/chenkaitao1112/EHHN.
arXiv:2607.01661v1 Announce Type: new Abstract: Multi-agent systems are increasingly used for forecasting future events, as deliberation among multiple LLMs is believed to improve reasoning and calibration. Yet existing approaches overlook a critical design choice: what information each agent receives. When all agents are given identical evidence, deliberation collapses into herding rather than genuine belief revision, leaving multi-agent systems little better than a single agent. We identify this as a fundamental gap and propose designed information asymmetry to close it: by partitioning evidence into shared public and disjoint private subsets, each agent holds exclusive knowledge that can only reach others through deliberation. We theoretically show that this decomposition reduces inter-agent error correlation, and instantiate it in InfoDelphi, a framework combining relevance-aware evidence routing, rationale-based iterative deliberation, and confidence-weighted aggregation. On PolyGym, a benchmark of 375 binary forecasting questions derived from real-world prediction markets, InfoDelphi outperforms the strongest single-agent and multi-agent baselines by 12--18% in Brier score and 4--8 percentage points in accuracy. More detailed experiments confirm that removing information asymmetry eliminates most deliberation gains, establishing diversity of input as the key enabler of effective multi-agent reasoning.
arXiv:2607.01662v1 Announce Type: new Abstract: Numerical modeling of aortic valve dynamics is essential for understanding the complex fluid-structure interaction (FSI) governing valve biomechanics in health and disease. Immersed methods provide a flexible computational framework for simulating the large deformations of valve leaflets and associated blood flow without requiring body-fitted meshes. Among these approaches, the Resistive Immersed Surface (RIS) and Immersed Boundary (IB) methods are widely used. However, systematic comparative analysis of these methods for realistic aortic valve simulations has not been performed. In this work, we compare a prescribed-kinematics RIS workflow implemented in SimVascular's svMultiPhysics solver with a fully coupled IB workflow using IBAMR for trileaflet and bicuspid aortic valve configurations. The RIS method represents the valve as a surface with prescribed kinematics embedded in the fluid domain and introduces a penalty force that drives the surrounding fluid velocity toward the prescribed leaflet velocity. This formulation reduces modeling complexity and provides useful hemodynamic predictions when representative leaflet kinematics are available. In contrast, the IB method models the leaflets as elastic structures fully immersed in the fluid domain and resolves leaflet deformation through fully coupled two-way FSI. The study focuses on the extent to which RIS reproduces bulk hemodynamic features and transvalvular pressure gradients. Results show that the RIS method captures the large-scale flow structures and predicts the mean transvalvular pressure gradient with a relative error within 15% of the fully coupled IB simulation, improving to within 5% when inlet boundary conditions are matched, while reducing computational cost by approximately 60%.
arXiv:2604.21254v3 Announce Type: replace Abstract: LLM architecture research generally aims to maximize model quality subject to fixed compute/latency budgets. However, many applications of interest such as edge and on-device deployment are further constrained by the model's memory footprint, thus motivating parameter-efficient architectures for language modeling. This paper describes a simple architecture that improves the parameter-efficiency of LLMs. Our architecture makes use of looped Transformers as a core primitive, which reuse Transformer layers across depth and are thus more parameter-efficient than ordinary (depth-matched) Transformers. We organize the looped Transformer into three blocks--begin, middle, and end blocks--where each block itself consists of multiple Transformer layers, and only the middle block is applied recurrently across depth. We augment the looped middle block with hyper-connections (Xie et al., 2026), which expand the residual stream into matrix-valued residual streams. Hyper-connections are applied only after each loop, and therefore add minimal new parameters and compute cost. Across various model scales, we find that our Hyper-Connected Looped Transformer (Hyperloop Transformer) is able to perform well compared to depth-matched Transformer and mHC Transformer baselines despite using approximately 50% fewer parameters. This performance persists through post-training weight quantization, thus positioning Hyperloop Transformers as an attractive architecture for memory-efficient language modeling.
arXiv:2607.01564v1 Announce Type: cross Abstract: Optimal encoding of classical data for quantum-assisted statistical inference is investigated from an information-theoretic perspective. We prove that the accuracy of any quantum-computing inference procedure is upper bounded by the maximal quantum leakage from the classical data through its quantum encoding, establishing leakage as a universal, task-agnostic quality measure for encoders. This demonstrates that the maximal quantum leakage is a universal measure of the quality of the encoding strategy for statistical inference as it only depends on the quantum encoding of the data and not the inference task itself. The optimal universal encoding strategy, i.e., an encoding strategy that maximizes the maximal quantum leakage, is proved to be attained by pure states. When there are enough qubits, basis encoding is proved to be universally optimal. However, when the dimension of the system is small, phase encoding is optimal. For the latter, any tight frame, any ensemble whose average state is the maximally mixed state, is in fact optimal. Within tight frames, equiangular tight frames (ETFs) are distinguished as the uniquely symmetric optimal encodings, i.e., they saturate the Welch lower bound on pairwise overlaps and possess a self-referential optimal measurement. Prominent special cases are the qubit trine, the regular simplex, and symmetric informationally complete positive operator-valued measures (SIC-POVMs), for which the ETF structure and explicit codeword constructions are provided. Numerical examples are presented to validate the theoretical predictions.
arXiv:2607.01713v1 Announce Type: cross Abstract: Machine-learning-accelerated materials discovery has yielded large numbers of computationally stable compounds, yet many remain experimentally unrealized, underscoring a persistent gap between prediction and synthesis. Here, we introduce a hierarchical screening framework that combines PBE-based thermodynamic stability, efficient dynamical-stability screening enabled by universal machine-learning interatomic potentials, and SCAN-based thermodynamic refinement. Applying this protocol to the 894 stable materials previously reported in Sci. Data 9, 302 (2022), we first curate 603 unique structures, of which only 298 remain thermodynamically stable on the complete PBE phase diagrams, demonstrating the critical role of competing phases in stability assessment. Dynamical screening then identifies 166 materials stable under both harmonic-phonon and finite-temperature molecular dynamics criteria, and SCAN phase diagrams further narrow this set to 109. Finally, by combining decomposition enthalpy with chemical-space completeness, we prioritize 25 candidates as high-confidence targets for experimental synthesis. This work provides a practical protocol for translating stability predictions into experimentally actionable synthesis targets, closing a key gap in machine-learning-driven materials discovery.
arXiv:2607.02182v1 Announce Type: new Abstract: Large language models (LLMs) exhibit remarkable reasoning capabilities, but their task-specific fine-tuning is notoriously plagued by overconfidence, severely hindering trustworthy deployment. We propose Data-Adaptive Lower-Rank Adaptation (DALorRA), a simple and effective variational Bayesian sparse framework that shifts the paradigm of uncertainty quantification from the dense parameter space to the lightweight rank level of low-rank adaptation (LoRA). With the insight that LoRA essentially aggregates multiple rank-one components that may provide superfluous model capacity, DALorRA imposes stochastic masking on rank dimensions, enabling Bayesian regularization of model capacity during training and ensemble-like calibration during inference. Extensive experiments demonstrate DALorRA's excellent calibration of LLMs without compromising reasoning accuracy.
arXiv:2602.03315v2 Announce Type: replace Abstract: Agent memory systems must accommodate continuously growing information while supporting efficient, context-aware retrieval for downstream tasks. Abstraction is essential for scaling agent memory, yet it often comes at the cost of specificity, obscuring the fine-grained details required for effective reasoning. We introduce Memora, a harmonic memory representation that structurally balances abstraction and specificity. Memora organizes information via its primary abstractions that index concrete memory values and consolidate related updates into unified memory entries, while cue anchors expand retrieval access across diverse aspects of the memory and connect related memories. Building on this structure, we employ a retrieval policy that actively exploits these memory connections to retrieve relevant information beyond direct semantic similarity. Theoretically, we show that standard Retrieval-Augmented Generation (RAG) and Knowledge Graph (KG)-based memory systems emerge as special cases of our framework. Empirically, Memora establishes a new state-of-the-art on the LoCoMo and LongMemEval benchmarks, demonstrating better retrieval relevance and reasoning effectiveness as memory scales.
arXiv:2602.05171v2 Announce Type: replace Abstract: The effect of Kolmogorov-size spherical particles on homogeneous and isotropic turbulence is investigated using particle-resolved direct numerical simulations at an unladen Taylor-scale Reynolds number of $150$. Four monodisperse suspensions of particles with identical diameter and volume fraction $10^{-3}$ are considered, while the particle-to-fluid density ratio varies between $100$ and $1500$ and the mass fraction between $0.1$ and $0.6$. As particle inertia increases, the energy spectrum departs from the canonical Kolmogorov $\kappa^{-5/3}$ scaling and approaches a peculiar regime with $\kappa^{-1}$. In this limit, the nonlinear energy transfer is strongly suppressed and the kinetic energy balance is dominated by the fluid-solid interaction and the viscous dissipation. Consistently, the second-order structure function shows logarithmic scaling at separations larger than the particle diameter, indicating velocity decorrelation. Increasing particle inertia promotes axial strain and vortex compression in the vicinity of the particles and enhances the particle-fluid relative velocity. Particle clustering is maximum when the Stokes number based on the Kolmogorov time scale is $O(1)$ and weakens as the density ratio and the Stokes number increase, with the volume and the population of the clusters decreasing when inertia is enhanced. When clustering occurs, particles preferentially sample regions of high strain and low vorticity.
arXiv:2607.02444v1 Announce Type: cross Abstract: We study stabilizer state testing and learning with limited coherent quantum memory. Here an algorithm sequentially receives copies of an unknown $n$-qubit state, but may keep only $k$ qubits of coherent quantum memory between measurements. With unrestricted memory, seminal work of Gross, Nezami and Walter showed how to test $n$-qubit stabilizer states using $6$ copies, which is dimension independent, unlike the learning complexity of $\Theta(n)$. We show that this testing-vs-learning separation is lost under memory constraints. More concretely we show that (1) The sample complexity of testing stabilizer states in the $k$-qubit memory framework is $\Theta(n-k)$. Our upper bound goes via a novel connection to the hidden shift problem and the lower bound is proven using a novel approach to average case bounds on likelihood ratios via combinatorics of the stochastic orthogonal group. (2) The sample complexity of learning stabilizer states with $k$ qubits of memory, in the non-adaptive framework, is $\Theta(n^2/k)$. As a further application of our techniques, we prove an exponential lower bound for purity testing even when the memory may be left coherent throughout the protocol. Our main results identify coherent quantum memory as the resource enabling the usual separation between stabilizer testing and learning. In particular, even with $k=0.99n$ qubits of memory, there is no constant-copy stabilizer tester; furthermore for $k=cn$ qubits of memory (for $0< c < 1$), stabilizer testing is as hard as learning, with both requiring $\Theta(n)$ copies.
arXiv:2512.07843v2 Announce Type: replace Abstract: Scaling inference-time computation has enabled Large Language Models (LLMs) to achieve strong reasoning performance, but their inherently sequential decoding incurs substantial latency, motivating parallelization of the generation process. However, existing parallel reasoning approaches suffer from performance degradation compared to their sequential counterparts, and often rely on specialized inference engines. We introduce ThreadWeaver, a framework for adaptive parallel reasoning that matches the accuracy of comparably sized sequential reasoning models while significantly reducing inference latency via three key innovations: 1) a two-stage parallel trajectory generator that produces high-quality parallel chain-of-thought data for supervised fine-tuning; 2) a trie-based rollout design that enables parallel reasoning on any off-the-shelf autoregressive inference engine; and 3) a parallelization-aware reinforcement learning framework that trains the model to balance reasoning accuracy with effective parallelization. Across six challenging math reasoning benchmarks, ThreadWeaver trained on top of Qwen3-8B achieves performance on par with cutting-edge sequential reasoning models (79.9% on AIME24 and 71.9% on average) while delivering up to 1.53x speedup in token latency, establishing a new Pareto frontier between accuracy and efficiency.
arXiv:2607.01764v1 Announce Type: new Abstract: Repository-level vulnerability reproduction is a demanding software engineering (SE) task: an agent must inspect a codebase, infer the input grammar that reaches a vulnerable path, construct a proof-of-conceptv(PoC), and verify that the crash disappears on the patched build. Recent LLM agents can often execute these steps when the approach is correct, yet they still fail by choosing the wrong strategy. This paper argues that strategy, rather than the full action trajectory, is the right learning unit for such SE agents: it is compact enough to optimize, concrete enough to guide execution, and stable enough to store and reuse across attempts. We present Mastermind, a dual-loop framework that separates transferable strategy learning from task-specific experience. A trainable planner learns reusable vulnerability-reproduction strategies through SFT and milestone-based GRPO, while an experience loop maintains task-local strategy records that guide subsequent attempts. The planner is trained independently of the executor, allowing strategy learning to improve multiple frozen executors without modifying their action-generation capability. We evaluate Mastermind on CyberGym using 260 training tasks and 200 held-out evaluation tasks. With GPT-5.5 as the frozen executor, Mastermind achieves an 84.5% pass rate, outperforming open-book PoC context (60.0%), Best-of-8 sampling (63.0%), and iterative improvement (77.0%). The same planner also improves GPT-5.4 mini and GLM~5.1 from 45.0% and 58.5% to 60.0% and 71.0%. These results demonstrate that learning high-level strategies is an effective and transferable mechanism for improving repository-scale SE agents.
arXiv:2607.01483v1 Announce Type: new Abstract: The ''Computational Life'' paper (Ag\"uera y Arcas et al., 2024) argues that paired interactions in a computational soup are an effective way to find self-replicators. In this work, aided by recent developments in self-replicator detection, we explore the alternate hypothesis that self-replicators can be found at least as easily using simple mutation random walks in program space. We also explore the claim that capping the maximum ''depth'' and ''width'' of the ancestry tree stops self-replicators from emerging, showing instead that it merely stops self-replicators from taking over the soup.
arXiv:2607.01492v1 Announce Type: new Abstract: Recent work has established a fundamental trilemma between Byzantine robustness, local differential privacy (LDP), and optimization error in distributed learning. We show that this trilemma does not universally extend to generalization error, but instead depends critically on the privacy regime. Specifically, in the high-noise regime (strong privacy), we prove that increasing privacy reduces the generalization error, i.e., there is no tension between robustness and privacy. In the low-noise regime (weaker privacy), however, the tension between robustness and privacy reappears and increasing privacy indeed degrades generalization. Our theory explains this surprising non-monotonic behavior of the generalization error via matching lower and upper bounds on the algorithmic stability of Byzantine-robust distributed learning under LDP constraints. We corroborate and further analyze these theoretical findings with empirical evaluations.
arXiv:2607.01686v1 Announce Type: new Abstract: Foundation models are routinely released to the public, yet the data recipes used to train them -- such as domain mixture weights that determine how different sources are sampled -- are rarely disclosed. This creates an access asymmetry: researchers study the resulting models but lack visibility into the training distribution that produces them. Prior works for inferring training data, such as membership inference, detect at the level of individual samples and thus cannot characterize the global composition of the training corpus. We introduce WARP, a framework that recovers a fine-tuned model's training mixtures directly from its released weights. WARP interpolates between the base and fine-tuned models using model merging, generating pseudo-checkpoints that approximate the missing training trajectory and expose a geometric footprint of the training data in the weight space. From these simulated footprints, WARP extracts geometric features and maps them to domain proportions using either a parameter-free softmax readout or an MLP projector trained on synthetic mixtures. In controlled experiments with BERT and GPT-2, WARP recovers domain mixtures with an average MAE as low as 0.046 and 0.104 respectively, outperforming membership inference and a variant with access to the true training trajectory.
arXiv:2306.01723v4 Announce Type: replace-cross Abstract: We present a polynomial-time quantum algorithm making a single query (in superposition) to a classical oracle, such that for every state $|\psi\rangle$ there exists a choice of oracle that makes the algorithm construct an exponentially close approximation of $|\psi\rangle$. Previous algorithms for this problem either used a linear number of queries and polynomial time, or a constant number of queries and polynomially many ancillae but no nontrivial bound on the runtime. As corollaries we do the following: - We simplify the proof that statePSPACE $\subseteq$ stateQIP (a quantum state analogue of PSPACE $\subseteq$ IP) and show that a constant number of rounds of interaction suffices. - We show that QAC$\mathsf{_f^0}$ lower bounds for constructing explicit states would imply breakthrough circuit lower bounds for computing explicit Boolean functions. - We prove that every $n$-qubit state can be constructed to within 0.01 error by an $O(2^n/n)$-size circuit over an appropriate finite gate set. More generally we give a size-error tradeoff which, by a counting argument, is optimal for any finite gate set.
arXiv:2311.17633v2 Announce Type: replace Abstract: Transformers have dominated empirical machine learning models of natural language processing. In this paper, we introduce basic concepts of Transformers and present key techniques that form the recent advances of these models. This includes a description of the standard Transformer architecture, a series of model refinements, and common applications. Given that Transformers and related deep learning techniques might be evolving in ways we have never seen, we cannot dive into all the model details or cover all the technical areas. Instead, we focus on just those concepts that are helpful for gaining a good understanding of Transformers and their variants. We also summarize the key ideas that impact this field, thereby yielding some insights into the strengths and limitations of these models.
arXiv:2604.08169v2 Announce Type: replace Abstract: Alignment in LLMs is more brittle than commonly assumed: misalignment can be induced by adversarial prompts, benign fine-tuning, emergent misalignment, and goal misgeneralization. Recent evidence suggests that some misalignment behaviors are encoded as linear structure in activation space, making it tractable via activation steering, which could be used as a lightweight runtime defense. We implement three methods: Steer-With-Fixed-Coefficient (SwFC), which applies uniform additive steering, and two novel projection-aware methods, Steer-to-Target-Projection (StTP) and Steer-to-Mirror-Projection (StMP), that use a logistic regression decision boundary to selectively intervene only on tokens whose activations fall below the threshold. We evaluate these methods on two threat models, dishonesty and dismissiveness, using malicious system prompts as a controlled proxy for misalignment. We conduct our experiments on two architectures (Llama-3.3-70B-Instruct and Qwen3.6-27B). All methods substantially recover alignment. StTP and StMP preserve general capabilities (MMLU, MT-Bench, AlpacaEval) better than uniform steering. Finally, we show that our honesty steering generalizes to out-of-distribution scenarios: a single honesty direction extracted from the aligned model significantly raises scores on the MASK benchmark, suppresses deception in multi-agent settings (Among Us), doubles the hidden-behavior discovery rate on AuditBench, and restores honesty in an emergently misaligned model
arXiv:2604.08766v2 Announce Type: replace Abstract: Scanpath prediction models forecast the sequence of fixations a person makes while searching a scene, and increasingly serve as the upstream perception layer for foveated rendering, intent inference, and gaze-driven assistive interfaces. Since eye-tracking data is expensive to collect, these models are routinely fine-tuned from public datasets or third-party pretrained weights, exposing them to training-time poisoning. We present the first backdoor attacks on multimodal scanpath prediction. This task differs from classification: its output is a continuous, variable-length sequence of fixations, opening new avenues of attack. A fixed-trajectory backdoor is easy to implant, but it clusters poisoned samples away from clean data, making it detectable. We instead design two backdoor attacks that condition the malicious supervision on each scene to keep triggered outputs diverse and plausible: a spatial misdirection attack that redirects the predicted search toward an attacker-chosen object instead of the queried one, and a duration inflation attack that lengthens the predicted search by inserting extra fixations while preserving correct localization. Our attacks succeed across visual, textual, and multimodal triggers, with duration inflation reaching up to 93.5% attack success from as few as 540 poisoned samples (2.5% of training data), and spatial misdirection redirecting the search in up to 61% of triggered inputs. We evaluate our attacks against five existing backdoor defenses, spanning fine-tuning, fine-pruning, neural attention distillation, contrastive learning, and trigger inversion, and show none removes the backdoor without degrading model output below the usable threshold. Our attacks generalize across models and datasets, showing scanpath prediction models are vulnerable to backdoor attacks through data poisoning, and that designing an effective defense remains an open problem.
arXiv:2607.01908v1 Announce Type: new Abstract: Large vision-language models (LVLMs) have achieved strong performance across many medical imaging tasks, yet their application to ultrasound remains limited due to its inherent complexity and variability. In this work, we revisit what is truly needed to enable real-world ultrasound understanding. Instead of introducing complex architectures or elaborate training strategies, we show that data scale and clinically faithful data alignment are the key factors. We construct a large-scale dataset of 1.5M real-world ultrasound examinations, containing 17.7M images, multi-organ coverage, and paired uncurated clinical reports. Crucially, we organize the data at the examination level, aligning multiple images with their corresponding reports to reflect real clinical workflows. We then fine-tune a standard LVLM using low-rank adaptation (LoRA) on this dataset without task-specific modifications. Surprisingly, this simple recipe already leads to strong performance across diverse ultrasound understanding tasks, outperforming prior methods designed with more complex pipelines. Beyond these results, we present model and data scaling analyses that provide insights into the role of scale in ultrasound LVLMs.
arXiv:2607.02322v1 Announce Type: new Abstract: Vision-Language-Action (VLA) models have shown remarkable promise in generalized robotic manipulation. However, their spatial generalization remains fragile. We argue that simply increasing the number of viewpoints is insufficient. Models often fall into the trap of Shortcut Learning, latching onto spurious correlations (e.g., fixed relative poses between objects or between the camera and robot base) rather than learning true spatial relationships. In this work, we propose a data-centric solution to enhance VLA spatial generalization. We utilize a dual-arm setup where one arm performs manipulation while the other serves as a mobile environmental camera. We systematically evaluate three data distribution patterns: Fixed, Multi-Fixed, and Moving Views. Our findings reveal that a hybrid strategy, combining continuous camera motion with diverse static viewpoints, yields the best performance by substantially reducing spurious correlations while maintaining training stability. Our experiments demonstrate that this strategy mitigates spurious correlations, enabling VLAs to generalize to unseen camera poses and object configurations where simply adding more static viewpoints fails. Crucially, we reveal that the susceptibility to shortcut learning and the struggle with spatial generalization are universal characteristics shared across diverse architectures. Consequently, all evaluated models (ACT, Diffusion, and VLA models including Pi0 and Gr00t) benefit significantly from our mixed data strategy.