AI conference papers

AI conference paper library

Search papers from 30+ top conferences, including NeurIPS, ICLR, ICML, CVPR, ACL and AAAI. Filter by keyword, then open any paper for its abstract, authors and links.

arXiv

Human-Aligned Procedural Level Generation Reinforcement Learning via Text-Level-Sketch Shared Representation

In-Chang Baek, Seoyoung Lee, Sung-Hyun Kim, Geumhwan Hwang, KyungJoong Kim

Human-aligned AI is a critical component of co-creativity, as it enables models to accurately interpret human intent and generate controllable outputs that align with design goals in collaborative content creation. This direction is especially relevant in procedural content generation via reinforcement learning (PCGRL), which is intended to serve as a tool for human designers. However, existing systems often fall short of exhibiting human-centered behavior, limiting the practical utility of AI-driven generation tools in real-world design workflows. In this paper, we propose VIPCGRL (Vision-Instruction PCGRL), a novel deep reinforcement learning framework that incorporates three modalities-text, level, and sketches-to extend control modality and enhance human-likeness. We introduce a shared embedding space trained via quadruple contrastive learning across modalities and human-AI styles, and align the policy using an auxiliary reward based on embedding similarity. Experimental results show that VIPCGRL outperforms existing baselines in human-likeness, as validated by both quantitative metrics and human evaluations. The code and dataset are available at https://github.com/bic4907/VIPCGRL.

arXiv

Unsupervised Deep Learning for Inverse Problems in Computed Tomography

Laura Hellwege, Johann Christopher Engster, Moritz Schaar, Thorsten M. Buzug, Maik Stille

Assume you encounter an inverse problem that shall be solved for a large number of data, but no ground-truth data is available. To emulate this, in this study we assume it is unknown how to solve the imaging problem of Computed Tomography. We introduce an unsupervised deep learning framework that leverages the inherent similarities between iterative reconstruction, Deep Image Prior (DIP), and unrolled optimization schemes. Our specific contribution is a training framework for amortized reconstruction: After training on a dataset without any image-domain ground truth, reconstruction of an unseen scan reduces to a single network forward pass. We demonstrate the feasibility of reconstructing images from measurement data by pure network inference, without additional gradient steps for unseen samples. Our method is evaluated on the two-dimensional 2DeteCT dataset. Within a controlled, geometry-matched benchmark, our reconstructions are competitive with, or better than, filtered back-projection, maximum-likelihood reconstruction, and a supervised network of identical architecture. Compared to a per-image DIP baseline, our method reaches similar quality while replacing the costly per-instance optimization with a single forward pass, yielding a speed-up of about four orders of magnitude. This makes it a promising candidate for time-critical imaging applications. Future work will address multi-dataset adaptability, counter-measures against over-smoothing, advanced uncertainty quantification, and further medical-imaging inverse problems.

arXiv

KDFlow: A User-Friendly and Efficient Knowledge Distillation Framework for Large Language Models

Songming Zhang, Xue Zhang, Tong Zhang, Bojie Hu, Yufeng Chen, Jinan Xu

Knowledge distillation (KD) is an essential technique to compress large language models (LLMs) into smaller ones. However, despite the distinct roles of the student model and the teacher model in KD, most existing frameworks still use a homogeneous training backend (e.g., FSDP and DeepSpeed) for both models, leading to suboptimal training efficiency. In this paper, we present a novel framework for LLM distillation, termed \textbf{KDFlow}, which features a decoupled architecture and employs SGLang for teacher inference. By bridging the training efficiency of FSDP2 and the inference efficiency of SGLang, KDFlow achieves full utilization of both advantages in a unified system. Moreover, instead of transferring full logits across different processes, our framework only transmits the teacher's hidden states using zero-copy data transfer and recomputes the logits on the student side, effectively balancing the communication cost and KD performance. Furthermore, our framework supports both off-policy and on-policy distillation and incorporates KD algorithms for cross-tokenizer KD through highly extensible and user-friendly APIs. Experiments show that KDFlow can achieve \textbf{1.44$\times$ to 6.36$\times$} speedup compared to current KD frameworks, enabling researchers to rapidly prototype and scale LLM distillation with minimal engineering overhead. Code is available at: https://github.com/songmzhang/KDFlow

arXiv

Jailbreak Foundry: From Papers to Runnable Attacks for Reproducible Benchmarking

Zhicheng Fang, Jingjie Zheng, Chenxu Fu, Wei Xu

Jailbreak techniques for large language models (LLMs) evolve faster than benchmarks, making robustness estimates stale and difficult to compare across papers due to drift in datasets, harnesses, and judging protocols. We introduce JAILBREAK FOUNDRY (JBF), a system that addresses this gap via a multi-agent workflow to translate jailbreak papers into executable modules for immediate evaluation within a unified harness. JBF features three core components: (i) JBF-LIB for shared contracts and reusable utilities; (ii) JBF-FORGE for the multi-agent paper-to-module translation; and (iii) JBF-EVAL for standardizing evaluations. Across 30 reproduced attacks, JBF achieves high fidelity with a mean (reproduced-reported) attack success rate (ASR) deviation of +0.26 percentage points. By leveraging shared infrastructure, JBF reduces attack-specific implementation code by more than half relative to original repositories and achieves an 82.5% mean reused-code ratio. This system enables a standardized AdvBench evaluation of all 30 attacks across 10 victim models using a consistent GPT-4o judge. By automating both attack integration and standardized evaluation, JBF offers a scalable solution for creating living benchmarks that keep pace with the rapidly shifting security landscape.

arXiv

Mechanistic Interpretability of Cognitive Complexity in LLMs via Linear Probing using Bloom's Taxonomy

Bianca Raimondi, Maurizio Gabbrielli

The black-box nature of Large Language Models necessitates novel evaluation frameworks that transcend surface-level performance metrics. This study investigates the internal neural representations of cognitive complexity using Bloom's Taxonomy as a hierarchical lens. By analyzing high-dimensional activation vectors from different LLMs, we probe whether different cognitive levels, ranging from basic recall (Remember) to abstract synthesis (Create), are linearly separable within the model's residual streams. Our results demonstrate that linear classifiers achieve approximately 95% mean accuracy across all Bloom levels, providing strong evidence that cognitive level is encoded in a linearly accessible subspace of the model's representations. These findings provide evidence that the model resolves the cognitive difficulty of a prompt early in the forward pass, with representations becoming increasingly separable across layers.

arXiv

Speculative Decoding with a Speculative Vocabulary

Miles Williams, Young D. Kwon, Rui Li, Alexandros Kouris, Stylianos I. Venieris

Speculative decoding has rapidly emerged as a leading approach for accelerating language model (LM) inference, as it offers substantial speedups while yielding identical outputs. This relies upon a small draft model, tasked with predicting the outputs of the target model. State-of-the-art speculative decoding methods use a draft model comprising a single decoder layer and output embedding matrix, with the latter dominating drafting time for the latest LMs. Recent work has sought to address this output distribution bottleneck by reducing the vocabulary of the draft model. While this can improve throughput, it compromises speculation effectiveness when the target token is out-of-vocabulary. In this paper, we argue for vocabulary speculation as an alternative to a reduced vocabulary. We propose SpecVocab, an efficient and effective method that selects a vocabulary subset per decoding step. Across a variety of tasks, we show that SpecVocab can achieve a higher acceptance length than state-of-the-art speculative decoding method, EAGLE-3. Notably, this yields up to an 8.1% increase in average throughput over EAGLE-3.

arXiv

Large-Scale Terminal Agentic Trajectory Generation from Dockerized Environments

Siwei Wu, Yizhi Li, Yuyang Song, Wei Zhang, Yang Wang, Riza Batista-Navarro, Xian Yang, Mingjie Tang, Bryan Dai, Jian Yang, Chenghua Lin

Training agentic models for terminal-based tasks critically depends on high-quality terminal trajectories that capture realistic long-horizon interactions across diverse domains. However, constructing such data at scale remains challenging due to two key requirements: \textbf{\emph{Executability}}, since each instance requires a suitable and often distinct Docker environment; and \textbf{\emph{Verifiability}}, because heterogeneous task outputs preclude unified, standardized verification. To address these challenges, we propose \textbf{TerminalTraj}, a scalable pipeline that (i) filters high-quality repositories to construct Dockerized execution environments, (ii) generates Docker-aligned task instances, and (iii) synthesizes agent trajectories with executable validation code. Using TerminalTraj, we curate 32K Docker images and generate 50,733 verified terminal trajectories across eight domains. Models trained on this data with the Qwen2.5-Coder backbone achieve consistent performance improvements on TerminalBench (TB), with gains of up to 20\% on TB~1.0 and 10\% on TB~2.0 over their respective backbones. Notably, \textbf{TerminalTraj-32B} achieves strong performance among models with fewer than 100B parameters, reaching 35.30\% on TB~1.0 and 22.00\% on TB~2.0, and demonstrates improved test-time scaling behavior. All code and data are available at https://github.com/Wusiwei0410/TerminalTraj.

arXiv

Discovering Generalizable Governing Equations for Graph Dynamical Systems with Interpretable Neural Networks

Riccardo Cappi, Paolo Frazzetto, Nicolò Navarin, Alessandro Sperduti

The discovery of symbolic governing equations is a central goal in science; yet, it remains challenging particularly for graph dynamical systems, where the network topology further shapes the system behavior. While artificial intelligence offers powerful tools for modeling these dynamics, the field lacks a rigorous comparative benchmark to assess the true scientific utility of the discovered laws. To address this challenge, this work proposes a novel evaluation pipeline designed to rigorously assess state-of-the-art symbolic regression models for graph equation discovery. Moving beyond simple fitting metrics, this framework evaluates discovered laws based on their long-term trajectory stability and, critically, their out-of-distribution generalization to unseen graph topologies. We benchmark established methods, including sparse regression and MLP-based architectures, and introduce the Graph Kolmogorov-Arnold Network-ODE (GKAN-ODE) model, a novel adaptation of KANs explicitly tailored for this domain, augmented by hyperparameter-free multiplicative nodes and a new Spline-Wise symbolic regression algorithm. Across a suite of synthetic and real-world graph dynamical systems, we numerically demonstrate through extensive experiments that neural-based approaches, particularly the GKAN-ODE model, recover exact ground-truth equations and achieve trajectory errors up to two orders of magnitude lower than the baseline methods on out-of-distribution test graphs.

arXiv

Label-Free Concept Drift Assessment for Reliable AI in Emerging Wireless Applications

Athanasios Tziouvaras, Carolina Fortuna, George Floros, Kostas Kolomvatsos, Panagiotis Sarigiannidis, Marko Grobelnik, Blaž Bertalanič

Machine learning models deployed in non-stationary environments degrade silently, since as the input distribution drifts their accuracy decays without an error signal and without labels to reveal it. Sustaining reliable AI therefore requires a concept-drift detector that acts as an external observer of the deployed model, monitoring it using unlabeled operational data alone, so that an MLOps actuator triggers retraining and redeployment only when it is warranted. This paper contributes two concept drift detectors, namely Confidence-Filtered Pseudo-Label Transfer (CFPT) and TabAutoDrift, which combine representation learning with statistical testing to compute an expected utility score that signals whether a deployed model should be retrained, without requiring ground-truth labels after deployment. The detectors are evaluated on two emerging, label-scarce wireless application domains in which post deployment ground truth is effectively unavailable, namely outdoor fingerprinting-based localization and link-anomaly detection. They outperform the classical detectors ADWIN, DDM, and CUSUM, attaining a drift-detection F1-score between 0.88 and 0.94 in the fingerprinting use case and between 0.80 and 1.00 in the link-anomaly use case, up to 0.24 higher than the strongest classical detector. Interpreted as reliability decisions, this precision indicates that the proposed detectors signal retraining more dependably.

arXiv

Boosted Enhanced Quantile Regression Neural Networks with Spatiotemporal Permutation Entropy for Complex System Prognostics

David J Poland

This paper presents an integrative prognostic framework that combines Spatiotemporal Permutation Entropy (STPE), Boosted Enhanced Quantile Regression Neural Networks (B-EQRNNs), Gated Temporal Attention, a Spiking Neural Network (SNN) refinement stage, and a Temporal Fusion Transformer (TFT) classifier. The motivation is long-horizon fault prediction in distributed industrial electronic systems, where single-sensor or point-estimate models can miss weak spatially propagating degradation signatures and provide limited uncertainty information. The proposed pipeline first converts 70-channel sensor streams into multiscale STPE descriptors, then learns conditional quantile representations and attention-weighted temporal context before final Normal/Abnormal classification. Evaluation is reported on a nine-system industrial electronic-sensor dataset with 48-, 90-, and 168-hour prediction horizons. The comparison includes a tree-based LightGBM baseline and modern sequence baselines available under the same preprocessing protocol, including LSTM, Autoformer, and TCN models. The full pipeline reaches 81.17% accuracy at the 168-hour horizon and is evaluated with component ablations, computational-cost analysis, and an explicit reproducibility protocol. The contribution is therefore framed as a validated hybrid architecture for uncertainty-aware spatiotemporal prognostics rather than as a new standalone learning theory.

arXiv

Rethinking Reward Signals in Video GRPO: When Scores Become Targets

Rui Li, Yuanzhi Liang, Ziqi Ni, Haibin Huang, Chi Zhang, Xuelong Li

Group Relative Policy Optimization (GRPO) enables stable and preference-oriented updates via group-wise comparisons for post-training video generation. However, GRPO directly optimizes reward-induced advantages. Under sustained optimization, the reward score can lose fidelity as a proxy for true video quality, consistent with the phenomenon described by Goodhart's Law. This leads to two recurring issues: (i) shortcut-driven optimization under composite objectives and (ii) reward saturation within prompt groups. To address these issues, we introduce TaRoS, a Target-Robust Reward Signaling framework for Video generation GRPO. TaRoS leverages component level performance assessment together with intra-group sparsity to organize multi-aspect rewards towards optimization objectives. In addition, it adaptively downweights components that exhibit saturation, thereby preserving effective optimization directions and mitigating redundancy. This maintains meaningful optimization directions and preserves within-group ranking separation, thereby preventing reward hacking and leading to more reliable policy updates. Extensive experiments show consistent improvements in visual fidelity, motion coherence, and text-video alignment over strong baselines.

arXiv

MVI-Bench: A Comprehensive Benchmark for Evaluating Robustness to Misleading Visual Inputs in LVLMs

Huiyi Chen, Jiawei Peng, Dehai Min, Changchang Sun, Kaijie Chen, Yan Yan, Xu Yang, Lu Cheng

Evaluating the robustness of Large Vision-Language Models (LVLMs) is essential for their continued development and responsible deployment in real-world applications. However, existing robustness benchmarks typically focus on hallucination or misleading textual inputs, while largely overlooking the equally critical challenge posed by misleading visual inputs in assessing visual understanding. To fill this important gap, we introduce MVI-Bench, the first comprehensive benchmark specially designed for evaluating how Misleading Visual Inputs undermine the robustness of LVLMs. Grounded in fundamental visual primitives, the design of MVI-Bench centers on three hierarchical levels of misleading visual inputs: Visual Concept, Visual Attribute, and Visual Relationship. Using this taxonomy, we curate six representative categories and compile 1,248 expertly annotated VQA instances. To facilitate fine-grained robustness evaluation, we further introduce MVI-Sensitivity, a novel metric that characterizes LVLM robustness at a granular level. Empirical results across 18 state-of-the-art LVLMs uncover pronounced vulnerabilities to misleading visual inputs, and our in-depth analyses on MVI-Bench provide actionable insights that can guide the development of more reliable and robust LVLMs. The benchmark and codebase can be accessed at https://github.com/chenyil6/MVI-Bench.

arXiv

Be Tangential to Manifold: Discovering Riemannian Metric for Diffusion Models

Shinnosuke Saito, Takashi Matsubara

Diffusion models are powerful deep generative models, but unlike classical models, they lack an explicit low-dimensional latent space that parameterizes the data manifold. This absence makes it difficult to perform manifold-aware operations, such as geometrically faithful interpolation or conditional guidance that respects the learned manifold. We propose a training-free Riemannian metric on the noise space, derived from the Jacobian of the score function. The key insight is that the spectral structure of this Jacobian separates tangent and normal directions of the data manifold; our metric leverages this separation to encourage paths to stay tangential to the manifold rather than drift toward high-density regions. To validate that our metric faithfully captures the manifold geometry, we examine it from two complementary angles. First, geodesics under our metric yield perceptually more natural interpolations than existing methods on synthetic, image, and video frame datasets. Second, the tangent-normal decomposition induced by our metric prevents classifier-free guidance from deviating off the manifold, improving generation quality while preserving text-image alignment.

arXiv

A Standardized Benchmark for Skeleton-Based Rehabilitation Assessment Using Deep Learning

Ali Ismail-Fawaz, Maxime Devanne, Stefano Berretti, Jonathan Weber, Germain Forestier

Automated assessment of human motion plays a vital role in rehabilitation, enabling objective evaluation of patient performance and progress. Unlike general human activity recognition, rehabilitation motion assessment focuses on analyzing the quality of movement within the same action class, requiring the detection of subtle deviations from ideal motion. Recent advances in deep learning and video-based skeleton extraction have opened new possibilities for accessible, scalable motion assessment using affordable devices such as smartphones or webcams. However, the field lacks standardized benchmarks, consistent evaluation protocols, and reproducible methodologies, limiting progress and comparability across studies. In this work, we address these gaps by (i) aggregating existing rehabilitation datasets into a unified archive called Rehab-Pile, (ii) proposing a general benchmarking framework for evaluating deep learning methods in this domain, and (iii) conducting extensive benchmarking of multiple architectures across classification and regression tasks. All datasets and implementations are released to the community to support transparency and reproducibility. This paper aims to establish a solid foundation for future research in automated rehabilitation assessment and foster the development of reliable, accessible, and personalized rehabilitation solutions. The datasets, source-code and results of this article are all publicly available.

arXiv

A Comprehensive Survey on Video Scene Parsing:Advances, Challenges, and Prospects

Guohuan Xie, Syed Ariff Syed Hesham, Wenya Guo, Bing Li, Ming-Ming Cheng, Guolei Sun, Yun Liu

Video Scene Parsing (VSP) studies dense video understanding, where every pixel in each frame must be segmented, each region must be named, and each object identity must remain coherent over time. This survey reviews recent progress in VSP across five tasks, spanning Video Semantic Segmentation (VSS), Video Instance Segmentation (VIS), Video Panoptic Segmentation (VPS), Video Tracking \& Segmentation (VTS), and Open-Vocabulary Video Segmentation (OVVS). We organize the literature as one architectural arc, running from hand-crafted motion and appearance cues through fully convolutional, attention-based and query-based designs to recent foundation-model approaches, and we trace how each family models temporal context, preserves identity, and balances accuracy against efficiency. We then compare the datasets, metrics and benchmark trends that shape current evaluation. Beyond cataloguing methods, we foreground the design trade-offs and recurring failure modes that cut across the field, namely temporal flicker, occlusion-induced identity switches, long-tail categories, and the annotation--capacity--latency tension. We close with open directions towards robust, efficient and open-world VSP systems.

arXiv

Acoustic Imaging for UAV Detection: Dense Beamformed Energy Maps and U-Net SELD

Belman Jahir Rodriguez, Sergio F. Chevtchenko, Marcelo Herrera Martinez, Yeshwanth Bethi, Saeed Afshar

We introduce a U-net model for 360° acoustic source localization formulated as a spherical semantic segmentation task. Rather than regressing discrete direction-of-arrival (DoA) angles, our model segments beamformed audio maps (azimuth & elevation) into regions of active sound presence. Using delay-and-sum (DAS) beamforming on a custom 24-microphone array, we generate signals aligned with drone GPS telemetry to create binary supervision masks. A modified U-Net, trained on frequency-domain representations of these maps, learns to identify spatially distributed source regions while addressing class imbalance via the Tversky loss. Because the network operates on beamformed energy maps, the approach is inherently array-independent and can adapt to different microphone configurations and can be transferred to different microphone configurations with minimal adaptation. The segmentation outputs are post-processed by computing centroids over activated regions, enabling robust DoA estimates. Our dataset includes real-world open-field recordings of a DJI Air 3 drone, synchronized with 360° video and flight logs across multiple dates and locations. Experimental results show that U-net generalizes across environments, providing improved angular precision, offering a new paradigm for dense spatial audio understanding beyond traditional Sound Source Localization (SSL). We additionally validate the same beamforming-plus-segmentation formulation on the DCASE 2019 TAU Spatial Sound Events benchmark, showing that the approach generalizes beyond drone acoustics to multiclass Sound Event Localization and Detection (SELD) scenarios.

arXiv

A Neuro-Symbolic Approach for Probabilistic Reasoning on Graph Data

Raffaele Pojer, Andrea Passerini, Kim G. Larsen, Manfred Jaeger

Graph neural networks (GNNs) excel at predictive tasks on graph-structured data but often lack the ability to incorporate symbolic domain knowledge and perform general reasoning. Relational Bayesian Networks (RBNs), in contrast, enable fully generative probabilistic modeling over graph-like structures and support rich symbolic knowledge and probabilistic inference. This paper presents a neuro-symbolic framework that seamlessly integrates GNNs into RBNs, combining the learning strength of GNNs with the flexible reasoning capabilities of RBNs. We develop two implementations of this integration: one compiles GNNs directly into the native RBN language, while the other maintains the GNN as an external component. Both approaches preserve the semantics and computational properties of GNNs while fully aligning with the RBN modeling paradigm. We also propose a maximum a posteriori (MAP) inference method for these neuro-symbolic models. To demonstrate the framework's versatility, we apply it to two distinct problems. First, we transform a GNN for node classification into a collective classification model that explicitly models homo- and heterophilic label patterns, substantially improving accuracy on both synthetic and real-world datasets. Second, we introduce a multi-objective network optimization problem in environmental planning, where MAP inference and continuous numeric relaxation support a complex decision-making task. Both applications include new publicly available benchmark datasets. This work introduces a powerful and coherent neuro-symbolic approach to graph data, bridging learning and reasoning in ways that enable novel applications and improved performance across diverse tasks.

arXiv

RAD: Retrieval High-quality Demonstrations to Enhance Decision-making

Lu Guo, Yixiang Shan, Zhengbang Zhu, Qifan Liang, Lichang Song, Ting Long, Weinan Zhang, Yi Chang

Offline reinforcement learning (RL) learns policies from fixed datasets, thereby avoiding costly or unsafe environment interactions. However, its reliance on finite static datasets inherently restricts the ability to generalize beyond the training distribution. Prior solutions based on synthetic data augmentation often fail to generalize to unseen scenarios in the (augmented) dataset. To address these challenges, we propose Retrieval High-quAlity Demonstrations (RAD) for decision-making, which innovatively introduces a retrieval mechanism into offline RL. Specifically, RAD retrieves high-return and reachable states from the offline dataset as target states, and leverages a generative model to generate sub-trajectories conditioned on these targets for planning. Since the targets are high-return states, once the agent reaches such a target, it can continue to obtain high returns by following the associated high-return actions, thereby improving policy generalization. Extensive experiments confirm that RAD achieves competitive or superior performance compared to baselines across diverse benchmarks, validating its effectiveness. Our code is available at https://github.com/LeahGL/RAD.

arXiv

Analysing Moral Bias in Finetuned LLMs through Mechanistic Interpretability

Bianca Raimondi, Daniela Dalbagno, Maurizio Gabbrielli

Large language models (LLMs) have been shown to internalize human-like biases during finetuning, yet the mechanisms by which these biases manifest remain unclear. In this work, we investigated whether the well-known Knobe effect, a moral bias in intentionality judgements, emerges in finetuned LLMs and whether it can be traced back to specific components of the model. We conducted a Layer-Patching analysis across 3 open-weights LLMs and demonstrated that the bias is not only learned during finetuning but also localized in a specific set of layers. Surprisingly, we found that patching activations from the corresponding pretrained model into just a few critical layers is sufficient to eliminate the effect. Our findings offer new evidence that social biases in LLMs can be interpreted, localized, and mitigated through targeted interventions, without the need for model retraining.

arXiv

Latent Fusion Jailbreak: Blending Harmful and Harmless Representations to Elicit Unsafe LLM Outputs

Wenpeng Xing, Bohan Yang, Mohan Li, Chunqiang Hu, Haitao Xu, Ningyu Zhang, Bo Lin, Meng Han

Safety-aligned large language models can still be manipulated through white-box interventions that modify their internal representations. We introduce Latent Fusion Jailbreak (LFJ), which works by pairing a harmful query with a structurally similar but benign counterpart, then interpolating their hidden states at carefully selected layers and token positions. Refusal-loss gradients determine exactly where to intervene, and we optimise layer-wise mixing coefficients using token-normalised compliance and refusal-suppression objectives. The edited prompt states propagate sequentially through the remaining transformer blocks. Across four safety benchmarks and five open-weight target models, LFJ reaches a macro-averaged attack success rate (ASR) of 94.13% under the white-box protocol we describe. Because LFJ directly accesses internal states, comparisons with prompt-only attacks serve as a descriptive reference rather than a matched evaluation. Dropping rejection sampling lowers ASR to 86.72%, whereas replacing the structured harmful-benign pairing with random pairing causes it to fall to 27.45%. We also design an LFJ-specific latent adversarial training procedure that, when the attack is re-optimised against the defended model, reduces ASR from 94.13% to 12.37%. This defence evaluation does not cover transfer to other attack types or preservation of benign utility.

Keep browsing

More papers from top AI conferences

30+ top conferences, by conference, year, and topic.