Machine Learning · Cross-listed from cs.CR
FedSpy-LLM: Towards Scalable and Generalizable Data Reconstruction Attacks from Gradients on LLMs
Syed Irfan Ali Meerza, Feiyi Wang and Jian Liu
Given the growing reliance on private data in training Large
Language Models (LLMs),
Federated Learning (FL) combined with Parameter-Efficient Fine-Tuning (PEFT) has garnered significant
attention for enhancing
privacy and efficiency. Despite FL's
privacy benefits, prior studies have shown that private data can still be extracted from shared gradients. However, these studies, mainly on full-parameter model training, are limited to reconstructing small batches, short input sequences, and specific model architectures, such as encoder-based or decoder-based models. The reconstruction quality becomes even worse when dealing with gradients from PEFT methods. To fully understand the practical attack surface of
federated LLMs, this paper proposes FedSpy-LLM, a scalable and generalizable data reconstruction attack designed to reconstruct training data with larger batch sizes and longer sequences while generalizing across diverse model architectures, even when PEFT methods are deployed for training. At the core of FedSpy-LLM is a novel gradient decomposition strategy that exploits the rank deficiency and subspace structure of gradients, enabling efficient
token extraction while preserving key signal components at scale. This approach further mitigates the reconstruction challenges introduced by PEFT's substantial null space, ensuring robustness across encoder-based, decoder-based, and encoder-decoder model architectures. Additionally, by iteratively aligning each
token's partial-sequence gradient with the full-sequence gradient, FedSpy-LLM ensures accurate
token ordering in reconstructed sequences.
Read the full abstract ▾
Machine Learning
AgentOpt v0.1 Technical Report: Client-Side Optimization for LLM-Based Agent
Wenyue Hua, Sripad Karne, Qian Xie et al. (+4 more)
AI
agents are increasingly deployed in real-world applications, including systems such as Manus, OpenClaw, and coding
agents. Existing research has primarily focused on server-side efficiency, proposing methods such as caching, speculative execution, traffic scheduling, and load balancing to reduce the cost of serving agentic workloads. However, as users increasingly construct
agents by composing local tools, remote APIs, and diverse models, an equally important optimization problem arises on the client side. Client-side optimization asks how developers should allocate the resources available to them, including model choice, local tools, and API budget across pipeline stages, subject to application-specific quality, cost, and latency constraints. Because these objectives depend on the task and deployment setting, they cannot be determined by server-side systems alone. We introduce AgentOpt, the first framework-agnostic Python package for client-side
agent optimization. We first study model selection, a high-impact optimization lever in multi-step
agent pipelines. Given a pipeline and a small evaluation set, the goal is to find the most cost-effective assignment of models to pipeline roles. This problem is consequential in practice: at matched accuracy, the cost gap between the best and worst model combinations can reach 13-32x in our experiments. To efficiently explore the exponentially growing combination space, AgentOpt implements ten search
algorithms, including UCB-E, UCB-E with Low-Rank Factorization, Arm Elimination, Epsilon-LUCB, Threshold Successive Elimination, and
Bayesian Optimization. Across four
benchmarks, UCB-E recovers near-optimal accuracy while reducing evaluation budget by 62-76% relative to brute-force search. Code and
benchmark results available at https://agentoptimizer.github.io/agentopt/.
Read the full abstract ▾
All authors Wenyue Hua, Sripad Karne, Qian Xie, Armaan Agrawal, Nikos Pagonas, Kostis Kaffes, Tianyi Peng
Machine Learning
TalkLoRA: Communication-Aware Mixture of Low-Rank Adaptation for Large Language Models
Lin Mu, Haiyang Wang, Li Ni et al. (+4 more)
Low-Rank Adaptation (LoRA) enables parameter-efficient fine-tuning of Large
Language Models (LLMs), and recent Mixture-of-Experts (MoE) extensions further enhance flexibility by dynamically combining multiple LoRA experts. However, existing MoE-augmented LoRA methods assume that experts operate independently, often leading to unstable routing, expert dominance. In this paper, we propose TalkLoRA, a communication-aware MoELoRA framework that relaxes this independence assumption by introducing expert-level communication prior to routing. TalkLoRA equips low-rank experts with a lightweight Talking Module that enables controlled information exchange across expert subspaces, producing a more robust global signal for routing. Theoretically, we show that expert communication smooths routing dynamics by mitigating perturbation amplification while strictly generalizing existing MoELoRA architectures. Empirically, TalkLoRA consistently outperforms vanilla LoRA and MoELoRA across diverse
language understanding and generation tasks, achieving higher parameter efficiency and more balanced expert routing under comparable parameter budgets. These results highlight structured expert communication as a principled and effective enhancement for MoE-based parameter-efficient adaptation. Code is available at https://github.com/why0129/TalkLoRA.
Read the full abstract ▾
All authors Lin Mu, Haiyang Wang, Li Ni, Lei Sang, Zhize Wu, Peiquan Jin, Yiwen Zhang
Machine Learning · Cross-listed from cs.CR
Adversarial Robustness of Time-Series Classification for Crystal Collimator Alignment
Xaver Fink, Borja Fernandez Adiego, Daniele Mirarchi et al. (+4 more)
In this paper, we analyze and improve the
adversarial robustness of a convolutional neural network (CNN) that assists
crystal-collimator
alignment at CERN's Large
Hadron Collider (LHC) by classifying a beam-loss monitor (BLM) time series during
crystal rotation. We formalize a local robustness property for this classifier under an
adversarial threat model based on real-world plausibility. Building on established parameterized input-transformation patterns used for transformation- and semantic-perturbation robustness, we instantiate a preprocessing-aware wrapper for our deployed time-series pipeline: we encode time-series normalization, padding constraints, and structured perturbations as a lightweight differentiable wrapper in front of the CNN, so that existing gradient-based robustness frameworks can operate on the deployed pipeline. For formal verification, data-dependent preprocessing such as per-window z-normalization introduces nonlinear operators that require verifier-specific abstractions. We therefore focus on attack-based robustness estimates and pipeline-checked validity by benchmarking robustness with the frameworks Foolbox and ART.
Adversarial fine-tuning of the resulting CNN improves robust accuracy by up to 18.6 % without degrading clean accuracy. Finally, we extend robustness on time-series data beyond single windows to sequence-level robustness for sliding-window classification, introduce
adversarial sequences as counterexamples to a temporal robustness requirement over full scans, and observe attack-induced misclassifications that persist across adjacent windows.
Read the full abstract ▾
All authors Xaver Fink, Borja Fernandez Adiego, Daniele Mirarchi, Eloise Matheson, Alvaro Garcia Gonzales, Gianmarco Ricci, Joost-Pieter Katoen
Machine Learning
Asymptotic-Preserving Neural Networks for Viscoelastic Parameter Identification in Multiscale Blood Flow Modeling
Giulia Bertaglia and Raffaella Fiamma Cabini
Mathematical models and numerical simulations offer a non-invasive way to explore cardiovascular phenomena, providing access to quantities that cannot be measured directly. In this study, we start with a one-dimensional multiscale blood flow model that describes the viscoelastic properties of arterial walls, and we focus on improving its practical applicability by addressing a major challenge: determining, in a reliable way, the viscoelastic parameters that control how arteries deform under pulsatile pressure. To achieve this, we employ Asymptotic-Preserving Neural Networks that embed the governing physical principles of the multiscale viscoelastic blood flow model within the learning procedure. This framework allows us to infer the viscoelastic parameters while simultaneously reconstructing the time-dependent
evolution of the state variables of blood vessels. With this approach, pressure waveforms are estimated from readily accessible patient-specific data, i.e., cross-sectional area and velocity measurements from Doppler ultrasound, in vascular segments where direct pressure measurements are not available. Different numerical simulations, conducted in both synthetic and patient-specific scenarios, show the effectiveness of the proposed methodology.
Read the full abstract ▾
Machine Learning · Cross-listed from stat.ML
Tight Convergence Rates for Online Distributed Linear Estimation with Adversarial Measurements
Nibedita Roy, Vishal Halder, Gugan Thoppe et al. (+4 more)
We study mean estimation of a random vector $X$ in a distributed parameter-server-worker setup. Worker $i$ observes samples of $a_i^\top X$, where $a_i^\top$ is the $i$th row of a known sensing
matrix $A$. The key challenges are
adversarial measurements and asynchrony: a fixed subset of workers may transmit corrupted measurements, and workers are activated asynchronously–only one is active at any time. In our previous work, we proposed a two-timescale $\ell_1$-minimization
algorithm and established asymptotic recovery under a null-space-property-like condition on $A$. In this work, we establish tight non-asymptotic convergence rates under the same null-space-property-like condition. We also identify relaxed conditions on $A$ under which exact recovery may fail but recovery of a projected component of $\mathbb{E}[X]$ remains possible. Overall, our results provide a unified finite-time characterization of robustness, identifiability, and statistical efficiency in distributed linear estimation with
adversarial workers, with implications for network
tomography and related distributed sensing problems.
Read the full abstract ▾
All authors Nibedita Roy, Vishal Halder, Gugan Thoppe, Alexandre Reiffers-Masson, Mihir Dhanakshirur, Naman, Alexandre Azor
Machine Learning · Cross-listed from cs.AI
Shoaib Sadiq Salehmohamed, Jinal Prashant Thakkar, Hansika Aredla et al. (+6 more)
Existing
hallucination detection methods for large
language models (LLMs) rely on external verification at inference time, requiring gold answers,
retrieval systems, or auxiliary judge models. We ask whether this external supervision can instead be distilled into the model's own representations during training, enabling
hallucination detection from internal activations alone at inference time. We introduce a weak supervision framework that combines three complementary grounding signals: substring matching, sentence
embedding similarity, and an LLM as a judge verdict to label generated responses as grounded or hallucinated without human annotation. Using this framework, we construct a 15000-sample
dataset from SQuAD v2 (10500 train/development samples and a separate 5000-sample test set), where each example pairs a LLaMA-2-7B generated answer with its full per-layer hidden states and structured
hallucination labels. We then train five probing classifiers: ProbeMLP (M0), LayerWiseMLP (M1), CrossLayerTransformer (M2), HierarchicalTransformer (M3), and CrossLayerAttentionTransformerV2 (M4), directly on these hidden states, treating external grounding signals as training-time supervision only. Our central hypothesis is that
hallucination detection signals can be distilled into
transformer representations, enabling internal detection without any external verification at inference time. Results support this hypothesis.
Transformer-based probes achieve the strongest discrimination, with M2 performing best on 5-fold average AUC/F1, and M3 performing best on both single-fold validation and held-out test evaluation. We also
benchmark inference efficiency: probe latency ranges from 0.15 to 5.62 ms (batched) and 1.55 to 6.66 ms (single sample), while end-to-end generation plus probe throughput remains approximately 0.231 queries per second, indicating negligible practical overhead.
Read the full abstract ▾
All authors Shoaib Sadiq Salehmohamed, Jinal Prashant Thakkar, Hansika Aredla, Shaik Mohammed Omar, Naveen Bera, Pulijala Sai Nikhila, Kondaguduru Abhiram, Shaik Gayaz Ali, Shalmali Ayachit
Machine Learning
RAGEN-2: Reasoning Collapse in Agentic RL
Zihan Wang, Chi Gui, Xing Jin et al. (+13 more)
RL training of multi-turn LLM
agents is inherently unstable, and
reasoning quality directly determines task performance.
Entropy is widely used to track
reasoning stability. However,
entropy only measures diversity within the same input, and cannot tell whether
reasoning actually responds to different inputs. In RAGEN-2, we find that even with stable
entropy, models can rely on fixed templates that look diverse but are input-agnostic. We call this template collapse, a failure mode invisible to
entropy and all existing metrics. To diagnose this failure, we decompose
reasoning quality into within-input diversity (
Entropy) and cross-input distinguishability (Mutual Information, MI), and introduce a family of mutual information proxies for online diagnosis. Across diverse tasks, mutual information correlates with final performance much more strongly than
entropy, making it a more reliable proxy for
reasoning quality. We further explain template collapse with a signal-to-noise ratio (SNR) mechanism. Low reward variance weakens task gradients, letting regularization terms dominate and erase cross-input
reasoning differences. To address this, we propose SNR-Aware Filtering to select high-signal
prompts per iteration using reward variance as a lightweight proxy. Across planning, math
reasoning, web
navigation, and code execution, the method consistently improves both input dependence and task performance.
Read the full abstract ▾
All authors Zihan Wang, Chi Gui, Xing Jin, Qineng Wang, Licheng Liu, Kangrui Wang, Shiqi Chen, Linjie Li, Zhengyuan Yang, Pingyue Zhang, Yiping Lu, Jiajun Wu, Li Fei-Fei, Lijuan Wang, Yejin Choi, Manling Li
Machine Learning
MO-RiskVAE: A Multi-Omics Variational Autoencoder for Survival Risk Modeling in Multiple MyelomaMO-RiskVAE
Zixuan Chen, Heng Zhang, YuPeng Qin et al. (+5 more)
Multimodal variational autoencoders (VAEs) have emerged as a powerful framework for survival risk modeling in multiple myeloma by integrating heterogeneous omics and clinical data. However, when trained under survival supervision, standard latent regularization strategies often fail to preserve prognostically relevant variation, leading to unstable or overly constrained representations. Despite numerous proposed variants, it remains unclear which aspects of latent design fundamentally govern performance in this setting. In this work, we conduct a controlled investigation of latent modeling choices for
multimodal survival prediction within a unified extension of the MyeVAE framework. By systematically isolating regularization scale, posterior geometry, and latent space structure under identical architectures and optimization protocols, we show that survival-driven training is primarily sensitive to the magnitude and structure of latent regularization rather than the specific divergence formulation. In particular, moderate relaxation of KL regularization consistently improves survival discrimination, while alternative divergence mechanisms such as MMD and HSIC provide limited benefit without appropriate scaling. We further demonstrate that structuring the latent space can improve
alignment between learned representations and survival risk gradients. A hybrid continuous–discrete formulation based on Gumbel–Softmax enhances global risk ordering in the continuous latent subspace, even though stable discrete subtype discovery does not emerge under survival supervision. Guided by these findings, we instantiate a robust
multimodal survival model, termed MO-RiskVAE, which consistently improves risk stratification over the original MyeVAE without introducing additional supervision or complex training heuristics.
Read the full abstract ▾
All authors Zixuan Chen, Heng Zhang, YuPeng Qin, WenPeng Xing, Qiang Wang, Da Wang, Changting Lin, Meng Han
Machine Learning
SMT-AD: a scalable quantum-inspired anomaly detection approach
Apimuk Sornsaeng, Si Min Chan, Wenxuan Zhang et al. (+4 more)
Quantum-inspired tensor networks
algorithms have shown to be effective and efficient models for machine learning tasks, including anomaly detection. Here, we propose a highly parallelizable
quantum-inspired approach which we call SMT-AD from Superposition of Multiresolution Tensors for Anomaly Detection. It is based upon the superposition of bond-dimension-1
matrix product operators to transform the input data with
Fourier-assisted feature
embedding, where the number of learnable parameters grows linearly with feature size,
embedding resolutions, and the number of additional components in the
matrix product operators structure. We demonstrate successful anomaly detection when applied to standard
datasets, including credit card transactions, and find that, even with minimal configurations, it achieves competitive performance against established anomaly detection baselines. Furthermore, it provides a straightforward way to reduce the weight of the model and even improve the performance by highlighting the most relevant input features.
Read the full abstract ▾
All authors Apimuk Sornsaeng, Si Min Chan, Wenxuan Zhang, Swee Liang Wong, Joshua Lim, Jonathan Pan, Dario Poletti
Machine Learning · Cross-listed from cs.GT
Incentive-Aware Multi-Fidelity Optimization for Generative Advertising in Large Language Models
Jiayuan Liu, Barry Wang, Jiarui Gan et al. (+4 more)
Generative advertising in large
language model (LLM) responses requires optimizing sponsorship configurations under two strict constraints: the strategic behavior of advertisers and the high cost of
stochastic generations. To address this, we propose the Incentive-Aware Multi-Fidelity Mechanism (IAMFM), a unified framework coupling Vickrey-Clarke-Groves (VCG) incentives with Multi-Fidelity Optimization to maximize expected social welfare. We compare two algorithmic instantiations (elimination-based and model-based), revealing their budget-dependent performance trade-offs. Crucially, to make VCG computationally feasible, we introduce Active Counterfactual Optimization, a "warm-start" approach that reuses optimization data for efficient payment calculation. We provide formal guarantees for approximate strategy-proofness and individual rationality, establishing a general approach for incentive-aligned, budget-constrained generative processes. Experiments demonstrate that IAMFM outperforms single-fidelity baselines across diverse budgets.
Read the full abstract ▾
All authors Jiayuan Liu, Barry Wang, Jiarui Gan, Tonghan Wang, Leon Xie, Mingyu Guo, Vincent Conitzer
Machine Learning
$S^3$: Stratified Scaling Search for Test-Time in Diffusion Language Models
Ahsan Bilal, Muhammad Ahmed Mohsin, Muhammad Umer et al. (+6 more)
Test-time scaling investigates whether a fixed
diffusion language model (DLM) can generate better outputs when given more inference compute, without additional training. However, naive best-of-$K$ sampling is fundamentally limited because it repeatedly draws from the same base
diffusion distribution, whose high-probability regions are often misaligned with high-quality outputs. We propose $S^3$ (Stratified Scaling Search), a classical verifier-guided search method that improves generation by reallocating compute during the denoising process rather than only at the final output stage. At each denoising step, $S^3$ expands multiple candidate trajectories, evaluates them with a lightweight reference-free verifier, and selectively resamples promising candidates while preserving diversity within the search frontier. This procedure effectively approximates a reward-tilted sampling distribution that favors higher-quality outputs while remaining anchored to the model prior. Experiments with LLaDA-8B-Instruct on MATH-500, GSM8K, ARC-Challenge, and TruthfulQA demonstrate that $S^3$ consistently improves performance across
benchmarks, achieving the largest gains on mathematical
reasoning tasks while leaving the underlying model and decoding schedule unchanged. These results show that classical search over denoising trajectories provides a practical mechanism for test-time scaling in DLMs.
Read the full abstract ▾
All authors Ahsan Bilal, Muhammad Ahmed Mohsin, Muhammad Umer, Asad Aali, Muhammad Usman Khanzada, Muhammad Usman Rafique, Zihao He, Emily Fox, Dean F. Hougen
Machine Learning
In-Place Test-Time Training
Guhao Feng, Shengjie Luo, Kai Hua et al. (+4 more)
The static “train then deploy" paradigm fundamentally limits Large
Language Models (LLMs) from dynamically adapting their weights in response to continuous streams of new information inherent in real-world tasks. Test-Time Training (TTT) offers a compelling alternative by updating a subset of model parameters (fast weights) at inference time, yet its potential in the current LLM ecosystem is hindered by critical barriers including architectural incompatibility, computational inefficiency and misaligned fast weight objectives for
language modeling. In this work, we introduce In-Place Test-Time Training (In-Place TTT), a framework that seamlessly endows LLMs with Test-Time Training ability. In-Place TTT treats the final projection
matrix of the ubiquitous MLP blocks as its adaptable fast weights, enabling a “drop-in" enhancement for LLMs without costly retraining from scratch. Furthermore, we replace TTT's generic reconstruction objective with a tailored, theoretically-grounded objective explicitly aligned with the Next-
Token-Prediction task governing autoregressive
language modeling. This principled objective, combined with an efficient chunk-wise update mechanism, results in a highly scalable
algorithm compatible with context parallelism. Extensive experiments validate our framework's effectiveness: as an in-place enhancement, it enables a 4B-parameter model to achieve superior performance on tasks with contexts up to 128k, and when pretrained from scratch, it consistently outperforms competitive TTT-related approaches. Ablation study results further provide deeper insights on our design choices. Collectively, our results establish In-Place TTT as a promising step towards a paradigm of continual learning in LLMs.
Read the full abstract ▾
All authors Guhao Feng, Shengjie Luo, Kai Hua, Ge Zhang, Di He, Wenhao Huang, Tianle Cai
Machine Learning
Unsupervised Topological Characterization of Slug, Churn and Annular Flow in Small-Diameter Vertical Pipes Using Euler Characteristic Surfaces
Brady Koenig, Sushovan Majhi, Atish Mitra et al. (+2 more)
Regime transitions in vertical gas-
liquid flow are read from mechanistic models or assigned by eye, and for churn flow neither route has produced a settled criterion. We ask how much an unsupervised topological description of flow
video can establish about them. The description is the Euler characteristic surface, a count of the connected gas and
liquid regions in each frame across morphological scales. Kernels derived from it are combined by unsupervised multiple kernel learning, and the trials are divided into contiguous groups along gas velocity. For 37 air-water trials in 2-in. tubing at one
liquid rate, with three viewing stations averaged, four groups are separated at 4.33, 7.51 and 11.70 m/s. Each division falls at or within one trial of a change in the visual evaluation of the same
videos; cuts placed at random would match as closely in 12% of cases. The divisions move between viewing stations and under resampling, a kernel built on gas velocity divides any sampled range into roughly equal parts and cannot locate a transition, and the data do not support a revised transition criterion. On independent images from an annulus, churn shows about three times the across-scale variance of slug. Within the band the published criteria assign to churn, the two upper cuts fall at graded changes in the rankings, consistent with a progression within churn; if confirmed by further data, they would subdivide the churn region.
Read the full abstract ▾
All authors Brady Koenig, Sushovan Majhi, Atish Mitra, Abigail Stein, Burt Todd
Machine Learning · Cross-listed from cs.CV
Reihaneh Zohrabi, Hosein Hasani, Akshita Gupta et al. (+3 more)
Large
vision-
language models can produce object
hallucinations in image descriptions, highlighting the need for effective detection and mitigation strategies. Prior work commonly relies on the model's
attention weights on visual
tokens as a detection signal. We reveal that coarse-grained
attention-based analysis is unreliable due to hidden confounders, specifically
token position and object repetition in a description. This leads to Simpson's paradox: the
attention trends reverse or disappear when statistics are aggregated. Based on this observation, we introduce HaloProbe, a
Bayesian framework that factorizes external description statistics and internal decoding signals to estimate
token-level
hallucination probabilities. HaloProbe uses balanced training to encourage learning from internal evidence and combines it with a learned prior over external features to recover the true posterior. While intervention-based mitigation methods often degrade utility or fluency by modifying models' internals, we use HaloProbe as an external scoring signal for non-invasive mitigation. Our experiments show that HaloProbe-guided decoding reduces
hallucinations more effectively than state-of-the-art intervention-based methods while preserving caption utility and linguistic fluency.
Read the full abstract ▾
All authors Reihaneh Zohrabi, Hosein Hasani, Akshita Gupta, Mahdieh Soleymani Baghshah, Anna Rohrbach, Marcus Rohrbach
Machine Learning · Cross-listed from cs.CV
The Character Error Vector: Decomposable errors for page-level OCR evaluation
Jonathan Bourne, Mwiza Simbeye and Joseph Nockels
The Character Error Rate (CER) is a key metric for evaluating the quality of Optical Character Recognition (OCR). However, this metric assumes that text has been perfectly parsed, which is often not the case. Under page-parsing errors, CER becomes undefined, limiting its use as a metric and making evaluating page-level OCR challenging, particularly when using data that do not share a labelling schema. We introduce the Character Error Vector (CEV), a bag-of-characters evaluator for OCR. The CEV can be decomposed into parsing and OCR, and interaction error components. This decomposability allows practitioners to focus on the part of the Document Understanding pipeline that will have the greatest impact on overall text extraction quality. The CEV can be implemented using a variety of methods, of which we demonstrate SpACER (Spatially Aware Character Error Rate) and a Character distribution method using the Jensen-Shannon Distance. We validate the CEV's performance against other metrics: first, the relationship with CER; then, parse quality; and finally, as a direct measure of page-level OCR quality. The validation process shows that the CEV is a valuable bridge between parsing metrics and local metrics like CER. We analyse a
dataset of archival newspapers made of degraded images with complex layouts and find that state-of-the-art end-to-end models are outperformed by more traditional pipeline approaches. Whilst the CEV requires character-level positioning for optimal triage, thresholding on easily available values can predict the main error source with an F1 of 0.91. We provide the CEV as part of a Python library to support Document understanding research.
Read the full abstract ▾
Machine Learning
Target Policy Optimization
Jean Kaddour
In RL, given a
prompt, we sample a group of completions from a model and score them. Two questions follow: which completions should gain probability mass, and how should the parameters move to realize that change? Standard policy-gradient methods answer both at once, so the update can overshoot or undershoot depending on the learning rate, clipping, and other optimizer choices. We introduce Target Policy Optimization (TPO), which separates the two questions. Given scored completions, TPO constructs a target distribution $q_i \propto p_i^{\,\mathrm{old}} \exp(u_i)$ and fits the policy to it by cross-
entropy. The loss gradient on sampled-completion logits is $p^\theta - q$, which vanishes once the policy matches the target. On tabular bandits,
transformer sequence tasks, and billion-parameter LLM RLVR, TPO matches PG, PPO, GRPO, and DG on easy tasks and substantially outperforms them under sparse reward. Code is available at https://github.com/JeanKaddour/tpo.
Read the full abstract ▾
Machine Learning
Toward Consistent World Models with Multi-Token Prediction and Latent Semantic Enhancement
Qimin Zhong, Hao Liao, Haiming Qin et al. (+4 more)
Whether Large
Language Models (LLMs) develop coherent internal world models remains a core debate. While conventional Next-
Token Prediction (NTP) focuses on one-step-ahead supervision, Multi-
Token Prediction (MTP) has shown promise in learning more structured representations. In this work, we provide a theoretical perspective analyzing the gradient inductive bias of MTP, supported by empirical evidence, showing that MTP promotes the convergence toward internal belief states by inducing representational contractivity via gradient coupling. However, we reveal that standard MTP often suffers from structural
hallucinations, where discrete
token supervision encourages illegal shortcuts in latent space that violate environmental constraints. To address this, we propose a novel method Latent Semantic Enhancement MTP (LSE-MTP), which anchors predictions to ground-truth hidden state trajectories. Experiments on synthetic graphs and real-world Manhattan Taxi Ride show that LSE-MTP effectively bridges the gap between discrete
tokens and continuous state representations, enhancing representation
alignment, reducing structural
hallucinations, and improving robustness to perturbations.
Read the full abstract ▾
All authors Qimin Zhong, Hao Liao, Haiming Qin, Mingyang Zhou, Rui Mao, Wei Chen, Naipeng Chao
Machine Learning · Cross-listed from quant-ph
Shot-based quantum encoding: a data-loading paradigm for quantum neural networks
Basil Kyriacou, Viktoria Patapovich, Maniraman Periyasamy et al. (+1 more)
Efficient data loading remains a bottleneck for near-term
quantum machine learning. Existing schemes (angle, amplitude, and basis encoding) either underuse the exponential
Hilbert-space capacity or require
circuit depths that exceed the coherence budgets of noisy intermediate-scale
quantum hardware. We introduce shot-based
quantum encoding (SBQE), a data
embedding strategy that distributes the hardware's native resource, shots, according to a data-dependent classical distribution over multiple initial
quantum states. By treating the shot counts as a learnable degree of freedom, SBQE produces a mixed-state representation whose expectation values are linear in the classical probabilities and can therefore be composed with nonlinear activation functions. We show that SBQE is structurally equivalent to a multilayer perceptron whose weights are realized by
quantum circuits, and we describe a hardware-compatible implementation protocol.
Benchmarks on three image
datasets, with 10 independent initializations per model, show that SBQE achieves 89.1% +- 0.9% test accuracy on Semeion (reducing error by 5.3% relative to amplitude encoding and matching a width-matched classical network), 80.95% +- 0.10% on Fashion MNIST (exceeding amplitude encoding by +2.0% and a linear multilayer perceptron by +1.3%), and 90.25% +- 0.18% on MNIST (exceeding amplitude encoding by 2.1 percentage points and the width-matched classical network by 0.3), all without any data-encoding gates.
Read the full abstract ▾
All authors Basil Kyriacou, Viktoria Patapovich, Maniraman Periyasamy, Alexey Melnikov
Machine Learning
Gym-Anything: Turn any Software into an Agent Environment
Pranjal Aggarwal, Graham Neubig and Sean Welleck
Computer-use
agents hold the promise of assisting in a wide range of digital economic activities. However, current research has largely focused on short-horizon tasks over a limited set of software with limited economic value, such as basic e-commerce and OS-configuration tasks. A key reason is that creating environments for complex software requires significant time and human effort, and therefore does not scale. To address this, we introduce Gym-Anything, a framework for converting any software into an interactive computer-use environment. We frame environment creation itself as a multi-
agent task: a coding
agent writes setup scripts, downloads real-world data, and configures the software, while producing evidence of correct setup. An independent audit
agent then verifies evidence for the environment setup against a quality checklist. Using a taxonomy of economically valuable occupations grounded in U.S. GDP data, we apply this pipeline to 200 software applications with broad occupational coverage. The result is CUA-World, a collection of over 10K long-horizon tasks spanning domains from medical science and astronomy to engineering and enterprise systems, each configured with realistic data along with train and test splits. CUA-World also includes CUA-World-Long, a challenging long-horizon
benchmark with tasks often requiring over 500 steps, far exceeding existing
benchmarks. Distilling successful trajectories from the training split into a 2B
vision-
language model outperforms models 2$\times$ its size. We also apply the same
auditing principle at test time: a separate VLM reviews completed trajectories and provides feedback on what remains, improving Gemini-3-Flash on CUA-World-Long from 11.5% to 14.0%. We release all code, infrastructure, and
benchmark data to facilitate future research in realistic computer-use
agents.
Read the full abstract ▾
Machine Learning · Cross-listed from stat.CO
A Large-Scale Empirical Comparison of Meta-Learners and Causal Forests for Heterogeneous Treatment Effect Estimation in Marketing Uplift Modeling
Aman Singh
Estimating Conditional Average Treatment Effects (CATE) at the individual level is central to precision marketing, yet systematic benchmarking of uplift modeling methods at industrial scale remains limited. We present UpliftBench, an empirical evaluation of four CATE
estimators: S-Learner, T-Learner, X-Learner (all with LightGBM base learners), and Causal Forest (EconML), applied to the Criteo Uplift v2.1
dataset comprising 13.98 million customer records. The near-random treatment assignment (propensity AUC = 0.509) provides strong internal validity for causal estimation. Evaluated via Qini coefficient and cumulative gain curves, the S-Learner achieves the highest Qini score of 0.376, with the top 20% of customers ranked by predicted CATE capturing 77.7% of all incremental conversions, a 3.9x improvement over random targeting. SHAP analysis identifies f8 as the dominant heterogeneous treatment effect (HTE) driver among the 12 anonymized
covariates. Causal Forest uncertainty quantification reveals that 1.9% of customers are confident persuadables (lower 95% CI > 0) and 0.1% are confident sleeping dogs (upper 95% CI < 0). Our results provide practitioners with evidence-based guidance on method selection for large-scale uplift modeling pipelines.
Read the full abstract ▾
Machine Learning
Low-Degree Learning from Localization Schemes
Gautam Chandrasekaran, Jason Gaitonde, Ankur Moitra et al. (+1 more)
The low-degree
algorithm is a cornerstone in computational learning theory. For product or Gaussian distributions, it shows that efficiently learnability follows from
Fourier concentration. However, a longstanding obstacle towards building such a theory over nonproduct distributions is that we can no longer rely on having closed-form expressions for a good basis of orthogonal functions. This hints that the analysis of the low-degree
algorithm might be brittle to the underlying distributional assumptions. We refute this intuition by showing that the low-degree
algorithm provably succeeds for a large class of graphical models under mild high-temperature conditions. As an application, for nearly every Ising model that is known to rapidly mix, we obtain learning
algorithms for every function class for which the
Fourier approach succeeds for product distributions (e.g. low-depth
circuits, halfspaces, intersections, monotone, among others) with no significant loss. Our framework is based on a strikingly clean connection with the coordinate-by-coordinate localization scheme of Chen and Eldan introduced for establishing rapid mixing in high-temperature models, as well as other analytical transference principles to product distributions.
Read the full abstract ▾
All authors Gautam Chandrasekaran, Jason Gaitonde, Ankur Moitra, Arsen Vasilyan
Machine Learning · Cross-listed from quant-ph
Pixel-Translation-Equivariant Quantum Convolutional Neural Networks via Fourier Multiplexers
Dmitry Chirkov and Igor Lobanov
Convolutional neural networks owe much of their success to hard-coding translation equivariance.
Quantum convolutional neural networks (QCNNs) have been proposed as near-term
quantum analogues, but the relevant notion of translation depends on the data encoding. For address/amplitude encodings such as FRQI, a pixel shift acts as modular addition on an index register, whereas many MERA-inspired QCNNs are equivariant only under cyclic permutations of physical
qubits. We formalize this mismatch and construct QCNN layers that commute exactly with the pixel cyclic shift (PCS)
symmetry induced by the encoding. Our main technical result is a constructive characterization of all PCS-equivariant
unitaries: conjugation by the
quantum Fourier transform (QFT) diagonalizes translations, so any PCS-equivariant layer is a
Fourier-mode multiplexer followed by an inverse QFT (IQFT). Building on this characterization, we introduce a deep PCS-QCNN with measurement-induced pooling, deferred conditioning, and inter-layer QFT cancellation. We also analyze trainability at random initialization and prove a lower bound on the expected squared gradient norm that remains constant in a depth-scaling regime, ruling out a depth-induced barren plateau in that sense.
Read the full abstract ▾
Machine Learning · Cross-listed from eess.SY
eVTOL Aircraft Energy Overhead Estimation under Conflict Resolution in High-Density Airspaces
Alex Zongo and Peng Wei
Electric vertical takeoff and landing (eVTOL) aircraft operating in high-density urban airspace must maintain safe separation through tactical conflict resolution, yet the energy cost of such maneuvers has not been systematically quantified. This paper investigates how conflict-resolution maneuvers under the Modified Voltage Potential (MVP)
algorithm affect eVTOL energy consumption. Using a physics-based power model integrated within a traffic simulation, we analyze approximately 71,767 en route sections within a sector, across traffic densities of 10-60 simultaneous aircraft. The main finding is that MVP-based deconfliction is energy-efficient: median energy overhead remains below 1.5% across all density levels, and the majority of en route flights within the sector incur negligible penalty. However, the distribution exhibits pronounced right-skewness, with tail cases reaching 44% overhead at the highest densities due to sustained multi-aircraft conflicts. The 95th percentile ranges from 3.84% to 5.3%, suggesting that a 4-5% reserve margin accommodates the vast majority of tactical deconfliction scenarios. To support operational planning, we develop a machine learning model that estimates energy overhead at mission initiation. Because conflict outcomes depend on future traffic interactions that cannot be known in advance, the model provides both point estimates and uncertainty bounds. These bounds are conservative; actual outcomes fall within the predicted range more often than the stated confidence level, making them suitable for safety-critical reserve planning. Together, these results validate MVP's suitability for energy-constrained eVTOL operations and provide quantitative guidance for reserve energy determination in Advanced Air Mobility.
Read the full abstract ▾
Machine Learning
A machine learning framework for uncovering stochastic nonlinear dynamics from noisy data
Matteo Bosso, Giovanni Franzese, Kushal Swamy et al. (+3 more)
Modeling real-world systems requires accounting for noise - whether it arises from unpredictable fluctuations in financial
markets, irregular rhythms in biological systems, or environmental variability in ecosystems. While the behavior of such systems can often be described by
stochastic differential equations, a central challenge is understanding how noise influences the inference of system parameters and dynamics from data. Traditional symbolic regression methods can uncover governing equations but typically ignore uncertainty. Conversely, Gaussian processes provide principled uncertainty quantification but offer little insight into the underlying dynamics. In this work, we bridge this gap with a hybrid symbolic regression-probabilistic machine learning framework that recovers the symbolic form of the governing equations while simultaneously inferring uncertainty in the system parameters. The framework combines deep symbolic regression with Gaussian process-based maximum likelihood estimation to separately model the deterministic dynamics and the noise structure, without requiring prior assumptions about their functional forms. We verify the approach on numerical
benchmarks, including harmonic, Duffing, and van der Pol oscillators, and validate it on an experimental system of coupled biological oscillators exhibiting synchronization, where the
algorithm successfully identifies both the symbolic and
stochastic components. The framework is data-efficient, requiring as few as 100-1000 data points, and robust to noise - demonstrating its broad potential in domains where uncertainty is intrinsic and both the structure and variability of dynamical systems must be understood.
Read the full abstract ▾
All authors Matteo Bosso, Giovanni Franzese, Kushal Swamy, Maarten Theulings, Alejandro M. Aragón, Farbod Alijani
Machine Learning · Cross-listed from cs.CL
Short Data, Long Context: Distilling Positional Knowledge in Transformers
Patrick Huber, Ernie Chang, Chinnadhurai Sankar et al. (+4 more)
Extending the context window of
language models typically requires expensive long-context pre-training, posing significant challenges for both training efficiency and data collection. In this paper, we present evidence that long-context
retrieval capabilities can be transferred to student models through logit-based knowledge
distillation, even when training exclusively on packed short-context samples within a long-context window. We provide comprehensive insights through the lens of Rotary Position
Embedding (RoPE) and establish three key findings. First, consistent with prior work, we show that phase-wise RoPE scaling, which maximizes rotational spectrum utilization at each training stage, also achieves the best long-context performance in knowledge
distillation setups. Second, we demonstrate that logit-based knowledge
distillation can directly enable positional information transfer. Using an experimental setup with packed repeated
token sequences, we trace the propagation of positional perturbations from query and key vectors through successive
transformer layers to output logits, revealing that positional information systematically influences the teacher's output distribution and, in turn, the
distillation signal received by the student model. Third, our analysis uncovers structured update patterns in the query state during long-context extension, with distinct parameter spans exhibiting strong sensitivity to long-context training.
Read the full abstract ▾
All authors Patrick Huber, Ernie Chang, Chinnadhurai Sankar, Rylan Conway, Igor Fedorov, Md Rifat Arefin, Adithya Sagar
Machine Learning · Cross-listed from math.OC
Value Mirror Descent for Reinforcement Learning
Zhichao Jia and Guanghui Lan
Value iteration-type methods have been extensively studied for computing a nearly optimal value function in reinforcement learning (RL). Under a generative sampling model, these methods can achieve sharper sample complexity than policy optimization approaches, particularly in their dependence on the discount factor. In practice, they are often employed for offline training. In this paper, we consider discounted Markov decision processes with state space S, action space A, discount factor $\gamma\in(0,1)$ and costs in $[0,1]$. We introduce a novel value optimization method, termed value mirror descent (VMD), which integrates mirror descent from convex optimization into the classical value iteration framework. In the deterministic setting with known transition kernels, we show that VMD converges linearly. For the
stochastic setting with a generative model, we develop a
stochastic variant, SVMD, which incorporates variance reduction commonly used in
stochastic value iteration-type methods. For RL problems with general convex regularizers, SVMD attains a near-optimal sample complexity of $\tilde{O}(|S||A|(1-\gamma)^{-3}\epsilon^{-2})$. Moreover, we establish that the Bregman divergence between the generated and optimal policies remains bounded throughout the iterations, even under the presence of model misspecification. This property is absent in existing
stochastic value iteration-type methods but is important for enabling effective online (continual) learning following offline training. Under a strongly convex regularizer, SVMD achieves sample complexity of $\tilde{O}(|S||A|(1-\gamma)^{-5}\epsilon^{-1})$, improving performance in the high-accuracy regime. Furthermore, we prove convergence of the generated policy to the optimal policy. Overall, the proposed method, its analysis, and the resulting guarantees, constitute new contributions to the RL and optimization literature.
Read the full abstract ▾
Machine Learning · Cross-listed from cs.DC
CodecSight: Leveraging Video Codec Signals for Efficient Streaming VLM Inference
Yulin Zou, Wenyan Chen, Yan Chen et al. (+6 more)
Continuous inference over concurrent
video streams imposes substantial compute and
memory demands on
vision-
language model (VLM) serving. Streaming inference uses sliding windows to maintain a bounded context of recent
video, but processing each window independently repeats visual encoding and large
language model (LLM) prefilling for similar and overlapping content. Existing optimizations provide limited coordination across these stages and often rely on model-specific training, profiling, or model-generated signals. We present CodecSight, a streaming VLM serving system that uses codec metadata as shared runtime guidance across visual encoding and LLM prefilling, without model-specific training or offline profiling. Codec-derived change signals guide patch
pruning before visual encoding, reducing both visual computation and the number of downstream visual
tokens. Codec-defined frame types guide selective key-value (KV) refresh across windows, while positional correction enables reuse of the remaining cached keys. Across three VLMs and four
video workloads, our vLLM-based implementation supports up to $3.3\times$ as many concurrent streams and achieves up to a $5.3\times$ speedup in average time-to-first-
token relative to the state-of-the-art baselines. It also reduces executed FLOPs by up to 93%, with a maximum task-quality decrease of 4.64 percentage points.
Read the full abstract ▾
All authors Yulin Zou, Wenyan Chen, Yan Chen, Anya Rajan, JooYoung Park, Shivaraman Nitin, Luo Tao, Francisco Romero, Dmitrii Ustiugov
Machine Learning · Cross-listed from stat.ML
Ensemble-Based Dirichlet Modeling for Predictive Uncertainty and Selective Classification
Courtney Franzen and Farhad Pourkamali-Anaraki
Neural network classifiers trained with cross-
entropy loss achieve strong predictive accuracy but lack the capability to provide inherent predictive uncertainty estimates, thus requiring external techniques to obtain these estimates. In addition, softmax scores for the true class can vary substantially across independent training runs, which limits the reliability of uncertainty-based decisions in downstream tasks. Evidential Deep Learning aims to address these limitations by producing uncertainty estimates in a single pass, but evidential training is highly sensitive to design choices including loss formulation, prior regularization, and activation functions. Therefore, this work introduces an alternative Dirichlet parameter estimation strategy by applying a method of moments
estimator to ensembles of softmax outputs, with an optional maximum-likelihood refinement step. This ensemble-based construction decouples uncertainty estimation from the fragile evidential loss design while also mitigating the variability of single-run cross-
entropy training, producing explicit Dirichlet predictive distributions. Across multiple
datasets, we show that the improved stability and predictive uncertainty behavior of these ensemble-derived Dirichlet estimates translate into stronger performance in downstream uncertainty-guided applications such as prediction confidence scoring and selective classification.
Read the full abstract ▾
Machine Learning
Gated-SwinRMT: Unifying Swin Windowed Attention with Retentive Manhattan Decay via Input-Dependent Gating
Dipan Maity, Suman Mondal and Arindam Roy
We introduce Gated-SwinRMT, a family of hybrid
vision transformers that combine the shifted-window
attention of the Swin
Transformer with the Manhattan-distance spatial decay of Retentive Networks (RMT), augmented by input-dependent gating. Self-
attention is decomposed into consecutive width-wise and height-wise retention passes within each shifted window, where per-head exponential decay masks provide a two-dimensional locality prior without learned positional biases. Two variants are proposed.Gated-SwinRMT-SWAT substitutes softmax with sigmoid activation, implements balanced ALiBi slopes with multiplicative post-activation spatial decay, and gates the value projection via SwiGLU; the Normalized output implicitly suppresses uninformative
attention scores. Gated-SwinRMT-Retention retains softmax-normalized retention with an additive log-space decay bias and incorporates an explicit G1 sigmoid gate – projected from the block input and applied after local context enhancement (LCE) but prior to the output projection~$W_O$ – to alleviate the low-rank $W_V \!\cdot\! W_O$ bottleneck and enable input-dependent suppression of attended outputs. We assess both variants on Mini-ImageNet ($224{\times}224$, 100 classes) and CIFAR-10 ($32{\times}32$, 10 classes) under identical training protocols, utilizing a single GPU due to resource limitations. At ${\approx}77$–$79$\,M parameters, Gated-SwinRMT-SWAT achieves $80.22\%$ and Gated-SwinRMT-Retention $78.20\%$ top-1 test accuracy on Mini-ImageNet, compared with $73.74\%$ for the RMT baseline. On CIFAR-10 – where small feature maps cause the adaptive windowing mechanism to collapse
attention to global scope – the accuracy advantage compresses from $+6.48$\,pp to $+0.56$\,pp.
Read the full abstract ▾