Negative / Null Result ReportOpen accessComputer Science
Andrei Liviu Nicolicioiu, Mohammad Pezeshki, Aaron Courville · 2026 · arXiv
On-policy self-distillation achieves strong pass@1 accuracy by using a single model as both teacher and student, with the teacher conditioned on a correct demonstration to provide dense token-level feedback. We show that this could come at a hidden cost: rollout diversity decreases and pass@k curves flatten (i.e., generating more rollouts fails to improve accuracy). We trace this to compounding biases in the design of self-distillation with sampled demonstrations. The teacher scores each student rollout while conditioned on a sampled correct rollout, channeling its feedback through the model's
View details →Negative / Null Result ReportOpen accessComputer Science
OFM Riaz Rahman Aranya, Kevin Desai · 2026 · arXiv
Synthetic data, an appealing alternative to extensive expert-annotated data for medical image segmentation, consistently fails to improve segmentation performance despite its visual realism. The reason being that synthetic and real medical images exist in different semantic feature spaces, creating a domain gap that current semi-supervised learning methods cannot bridge. We propose SRA-Seg, a framework explicitly designed to align synthetic and real feature distributions for medical image segmentation. SRA-Seg introduces a similarity-alignment (SA) loss using frozen DINOv2 embeddings to pull s
View details →Failed Experiment ReportOpen accessComputer Science
Chen-Rong Liu, Chuang Li, Runxia Tao et al. · 2026 · arXiv
Conventional noise analysis in atomic-ensemble sensing assumes a continuous-medium approximation, thereby treating the atomic system as a deterministic dielectric. Here, we demonstrate that this assumption breaks down due to the discrete, particulate nature of the ensemble, giving rise to an intrinsic "atomic granularity noise" (AGN) that fundamentally competes with the optical measurement noise (OMN, typically photon shot noise). By introducing a discrete-atom statistical framework, we derive a unified noise-scaling law governed by a single dimensionless resource ratio, $\mathcal{R} = \bar{N}
View details →Negative / Null Result ReportOpen accessEngineering
Sreyan Ghosh, Mohammad Sadegh Rasooli, Michael Levit et al. · 2024 · arXiv
Generative Error Correction (GEC) has emerged as a powerful post-processing method to enhance the performance of Automatic Speech Recognition (ASR) systems. However, we show that GEC models struggle to generalize beyond the specific types of errors encountered during training, limiting their ability to correct new, unseen errors at test time, particularly in out-of-domain (OOD) scenarios. This phenomenon amplifies with named entities (NEs), where, in addition to insufficient contextual information or knowledge about the NEs, novel NEs keep emerging. To address these issues, we propose DARAG (D
View details →Negative / Null Result ReportMedicine
Yang, Mueller, D'Andrea et al. · 2026 · Academic psychiatry : the journal of the American Association of Directors of Psychiatric Residency Training and the Association for Academic Psychiatry
Resident physicians experience high rates of depression, anxiety, burnout, and loneliness, yet few evidence-based interventions have been evaluated in this population. This randomized controlled pilot trial examined the feasibility,…
View details →Negative / Null Result ReportOpen accessComputer Science
Guowei Liu, Hongming Li, Yaning Guo et al. · 2026 · arXiv
Deploying large-scale MoE models presents challenges in memory capacity and bandwidth for expert activation. While Attention-FFN Disaggregation (AFD) has emerged as a potential architecture to decouple compute and memory resources, its performance boundaries compared to standard large-scale Expert Parallelism (EP) remain underexplored. In this paper, we conduct a systematic analysis of AFD by extending the roofline model to the communication level, correlating interconnect bandwidth, arithmetic intensity, and Hardware FLOPS Utilization (HFU). Our analysis reveals a dead zone on standard cluste
View details →Negative / Null Result ReportOpen accessComputer Science
Julian Coda-Forno, Zhuokai Zhao, Qiang Zhang et al. · 2025 · arXiv
Should LLM reasoning live in a separate module, or within a single model's forward pass and representational space? We study dual-architecture latent reasoning, where a fluent Base exchanges latent messages with a Coprocessor, and test two hypotheses aimed at improving latent communication over Liu et al. (2024): (H1) increase channel capacity; (H2) learn communication via joint finetuning. Under matched latent-token budgets on GPT-2 and Qwen-3, H2 is consistently strongest while H1 yields modest gains. A unified soft-embedding baseline, a single model with the same forward pass and shared rep
View details →Negative / Null Result ReportOpen accessComputer Science
Sepanta Zeighami, Cyrus Shahabi · 2024 · arXiv
While extremely useful (e.g., for COVID-19 forecasting and policy-making, urban mobility analysis and marketing, and obtaining business insights), location data collected from mobile devices often contain data from a biased population subset, with some communities over or underrepresented in the collected datasets. As a result, aggregate statistics calculated from such datasets (as is done by various companies including Safegraph, Google, and Facebook), while ignoring the bias, leads to an inaccurate representation of population statistics. Such statistics will not only be generally inaccurate
View details →Negative / Null Result ReportOpen accessComputer Science
Dong Xu, Jiantao Wu, Qihua Pan et al. · 2026 · arXiv
Drug-drug interaction (DDI) prediction is central to drug discovery and clinical development, particularly in the context of increasingly prevalent polypharmacy. Although existing computational methods achieve strong performance on standard benchmarks, they often fail to generalize to realistic deployment scenarios, where most candidate drug pairs involve previously unseen drugs and validated interactions are scarce. We demonstrate that proximity in the embedding spaces of prevailing molecule-centric DDI models does not reliably correspond to interaction labels, and that simply scaling up mode
View details →Negative / Null Result ReportOpen accessComputer Science
Romain Cosentino, Sarath Shekkizhar, Adam Earle et al. · 2026 · arXiv
Negotiation requires more than inferring what the other side wants: it requires using that information to make advantageous offers and counteroffers over multiple turns. We study whether large language model (LLM) agents do this in a controlled multi-attribute bargaining environment. We find that current LLM agents can model a counterparty's preferences, but do not reliably turn that knowledge into strategic bargaining. When given negotiating partner preference information, agents model it accurately and early in their reasoning traces, yet this does not reliably improve outcomes for the infor
View details →Negative / Null Result ReportOpen accessComputer Science
Mike Zhang, Ali Basirat, Desmond Elliott · 2026 · arXiv
Prior work establishes that controlled contrastiveness between self-generated responses from large language models, set via reward scores, improves downstream preference tuning in English. We extend this method to multiple languages and evaluate two models across a total of 14 high and low-resource languages on a diverse set of tasks. Our central finding is that cross-lingual contrastive preference tuning on self-generations (CroCo) transfers without language-specific preference annotation. A reward model trained on English preferences (atop a multilingual base) produces useful within-language
View details →Negative / Null Result ReportOpen accessComputer Science
Aaron Baier-Reinio, Hans De Sterck · 2020 · arXiv
We use neural ordinary differential equations to formulate a variant of the Transformer that is depth-adaptive in the sense that an input-dependent number of time steps is taken by the ordinary differential equation solver. Our goal in proposing the N-ODE Transformer is to investigate whether its depth-adaptivity may aid in overcoming some specific known theoretical limitations of the Transformer in handling nonlocal effects. Specifically, we consider the simple problem of determining the parity of a binary sequence, for which the standard Transformer has known limitations that can only be ove
View details →Negative / Null Result ReportMedicine
Escoffier, Hedhli, Campos-Juanatey et al. · 2026 · The French journal of urology
Artificial intelligence (AI) is increasingly used in surgery, but its role in reconstructive urology remains insufficiently studied. The aim of this work was to evaluate the theoretical knowledge of several AI platforms and compare their…
View details →Negative / Null Result ReportOpen accessComputer Science
Rui Xing, Qi Chai, Jie Ma et al. · 2026 · arXiv
Hate speech online targets individuals or groups based on identity attributes and spreads rapidly, posing serious social risks. Memes, which combine images and text, have emerged as a nuanced vehicle for disseminating hate speech, often relying on cultural knowledge for interpretation. However, existing multimodal hate speech datasets suffer from coarse-grained labeling and a lack of integration with surrounding discourse, leading to imprecise and incomplete assessments. To bridge this gap, we propose an agentic annotation framework that coordinates seven specialized agents to generate hierarc
View details →Negative / Null Result ReportOpen accessComputer Science
Zhiwei Jia, Xuanlin Li, Zhan Ling et al. · 2022 · arXiv
Generalization in deep reinforcement learning over unseen environment variations usually requires policy learning over a large set of diverse training variations. We empirically observe that an agent trained on many variations (a generalist) tends to learn faster at the beginning, yet its performance plateaus at a less optimal level for a long time. In contrast, an agent trained only on a few variations (a specialist) can often achieve high returns under a limited computational budget. To have the best of both worlds, we propose a novel generalist-specialist training framework. Specifically, w
View details →Negative / Null Result ReportOpen accessComputer Science
Jiamin Xu, Jacqueline Maasch, Kyra Gan · 2026 · arXiv
Online reinforcement learning (RL) relies on the Markov property for guaranteed performance, but real-world applications often lack well-defined states given raw observed variables. While causal RL has attracted growing interest, existing work typically assumes Markovian states are provided and focuses on using causality to accelerate learning, leaving a fundamental gap: \emph{given a longitudinal causal graph over observed variables, how does one construct MDP states that provably satisfy the Markov property?} We address this by providing a procedure that constructs a provably minimal state r
View details →Negative / Null Result ReportOpen accessComputer Science
Lily H. Zhang, Rajesh Ranganath · 2023 · arXiv
Methods which utilize the outputs or feature representations of predictive models have emerged as promising approaches for out-of-distribution (OOD) detection of image inputs. However, these methods struggle to detect OOD inputs that share nuisance values (e.g. background) with in-distribution inputs. The detection of shared-nuisance out-of-distribution (SN-OOD) inputs is particularly relevant in real-world applications, as anomalies and in-distribution inputs tend to be captured in the same settings during deployment. In this work, we provide a possible explanation for SN-OOD detection failur
View details →Negative / Null Result ReportOpen accessComputer Science
Zhaofeng Wu, Shiqi Wang, Boya Peng et al. · 2026 · arXiv
Modern language models demonstrate impressive coding capabilities in common programming languages (PLs), such as C++ and Python, but their performance in lower-resource PLs is often limited by training data availability. In principle, however, most programming skills are universal across PLs, so the capability acquired in one PL should transfer to others. In this work, we propose the task of zero-shot cross-programming-language transfer for code RL. We find that, for Llama-3.1, RL training for code generation in a source PL fails to improve, and sometimes even degrades, the performance on othe
View details →Negative / Null Result ReportOpen accessMathematics
Ziling Ma, Ángel López Oriona, Hernando Ombao et al. · 2026 · arXiv
We study adaptive pooling under predictive heterogeneity in high-dimensional multivariate time series forecasting, where global models improve statistical efficiency but may fail to capture heterogeneous predictive structure, while naive specialization can induce negative transfer. We formulate adaptive pooling as a statistical decision problem and propose a validation-driven framework that determines when and how specialization should be applied. Rather than grouping series based on representation similarity, we define partitions through out-of-sample predictive performance, thereby aligning
View details →Negative / Null Result ReportOpen accessMathematics
Rong Chen, Simone Giannerini, Greta Goracci et al. · 2025 · arXiv
We develop an estimation methodology for a factor model for high-dimensional matrix-valued time series, where common stochastic trends and common stationary factors can be present. We study, in particular, the estimation of (row and column) loading spaces, of the common stochastic trends and of the common stationary factors, and the row and column ranks thereof. In a set of (negative) preliminary results, we show that a projection-based technique fails to improve the rates of convergence compared to a "flattened" estimation technique which does not take into account the matrix nature of the da
View details →Negative / Null Result ReportMedicine
Yip, Xu, Liu et al. · 2026 · Asia-Pacific journal of ophthalmology (Philadelphia, Pa.)
To build a visual question answering (VQA) dataset for fine-tuning and evaluating vision-language models (VLMs) in myopic maculopathy (MM). Cross-sectional study. Colour fundus photographs (CFPs) from two publicly available datasets were…
View details →Negative / Null Result ReportOpen accessComputer Science
Philipp Spohn, Leander Girrbach, Zeynep Akata · 2026 · arXiv
Typical LLM responses tend to follow a default style, even though users often have distinct preferences regarding tone, verbosity, and formality that they do not explicitly state in their prompts. Evaluating whether personalization methods can adapt to these implicit preferences is challenging, since users typically provide prompts rather than reference responses, style preferences are not factually verifiable, and reference-free LLM judges may conflate personalization with general response quality. To address these challenges, we introduce the Arbitrary Preference Mapping (APM) benchmark, whi
View details →Negative / Null Result ReportOpen accessEngineering
Mengqi Wang, Zhan Liu, Zengrui Jin et al. · 2025 · arXiv
Diffusion-based large language models (DLLMs) have recently attracted growing interest as an alternative to autoregressive decoders. In this work, we present an empirical study on using the diffusion-based large language model LLaDA for automatic speech recognition (ASR). We first investigate its use as an external deliberation-based processing module for Whisper-LLaMA transcripts. By leveraging the bidirectional attention and denoising capabilities of LLaDA, we explore random masking, low-confidence masking, and semi-autoregressive strategies, showing that Whisper-LLaDA substantially reduces
View details →Negative / Null Result ReportOpen accessComputer Science
Minwu Kim, Anubhav Shrestha, Safal Shrestha et al. · 2025 · arXiv
Recent studies have shown that reinforcement learning with verifiable rewards (RLVR) enhances overall accuracy (pass@1) but often fails to improve capability (pass@k) of LLMs in reasoning tasks, while distillation can improve both. In this paper, we investigate the mechanisms behind these phenomena. First, we demonstrate that RLVR struggles to improve capability as it focuses on improving the accuracy of the easier questions to the detriment of the accuracy of the most difficult questions. Second, we show that RLVR does not merely increase the success probability for the easier questions, but
View details →Negative / Null Result ReportOpen accessComputer Science
Hongsin Lee, Hye Won Chung · 2026 · arXiv
Adversarial Distillation aims to enhance student robustness by guiding the student with a robust teacher's soft labels within the min-max adversarial training framework, yet its success is notoriously inconsistent: a more robust teacher often fails to improve, or even harms, the student's robust generalization. In this paper, we identify a key mechanism of this teacher dependency: the misalignment between the teacher's supervisory confidence and the student's representational limitations on a consistent subset of training data -- the Robustly Unlearnable Set. We present a theoretical framework
View details →Negative / Null Result ReportOpen accessComputer Science
Mujtaba Farhan, Maheep Chaudhary · 2026 · arXiv
Large language models (LLMs) have demonstrated remarkable reasoning abilities on mathematical and multi-hop planning tasks. The CoCoNuT (Chain of Continuous Thought) paradigm~\cite{hao2024coconut} extends this by enabling models to reason in latent space, exploring multiple reasoning paths simultaneously rather than committing to a single chain early on. However, we identify a limitation we term the \textbf{concept bottleneck}. At each reasoning pass, intermediate hidden states are overwritten, causing the model to lose critical facts computed in earlier steps as reasoning depth increases. We
View details →Negative / Null Result ReportOpen accessComputer Science
Anirudh Vemula, J. Andrew Bagnell, Maxim Likhachev · 2020 · arXiv
Given access to accurate dynamical models, modern planning approaches are effective in computing feasible and optimal plans for repetitive robotic tasks. However, it is difficult to model the true dynamics of the real world before execution, especially for tasks requiring interactions with objects whose parameters are unknown. A recent planning approach, CMAX, tackles this problem by adapting the planner online during execution to bias the resulting plans away from inaccurately modeled regions. CMAX, while being provably guaranteed to reach the goal, requires strong assumptions on the accuracy
View details →Negative / Null Result ReportOpen accessComputer Science
Yaowen Ye, Cassidy Laidlaw, Jacob Steinhardt · 2025 · arXiv
Language model (LM) post-training relies on two stages of human supervision: task demonstrations for supervised finetuning (SFT), followed by preference comparisons for reinforcement learning from human feedback (RLHF). As LMs become more capable, the tasks they are given become harder to supervise. Will post-training remain effective under unreliable supervision? To test this, we simulate unreliable demonstrations and comparison feedback using small LMs and time-constrained humans. We find that in the presence of unreliable supervision, SFT still retains some effectiveness, but DPO (a common
View details →Negative / Null Result ReportOpen accessComputer Science
Yiqian Yang, Tian Lan, Qianghuai Jia et al. · 2025 · arXiv
Effective deep search agents must not only access open-domain and domain-specific knowledge but also apply complex rules-such as legal clauses, medical manuals and tariff rules. These rules often feature vague boundaries and implicit logic relationships, making precise application challenging for agents. However, this critical capability is largely overlooked by current agent benchmarks. To fill this gap, we introduce HSCodeComp, the first realistic, expert-level e-commerce benchmark designed to evaluate deep search agents in hierarchical rule application. In this task, the deep reasoning proc
View details →Negative / Null Result ReportOpen accessEngineering
Jiaqi Wu, Jingyi Yuan, Yang Weng et al. · 2025 · arXiv
Power system voltage regulation is crucial to maintain power quality while integrating intermittent renewable resources in distribution grids. However, the system model on the grid edge is often unknown, making it difficult to model physical equations for optimal control. Therefore, previous work proposes structured data-driven methods like input convex neural networks (ICNN) for "optimal" control without relying on a physical model. While ICNNs offer theoretical guarantees based on restrictive assumptions of non-negative neural network parameters, can one improve the approximation power with
View details →