arXiv subjects

Zeyu Feng

Publications and source records attributed to Zeyu Feng.

14 recordsLinked to original sources

arXiv2026

PsychJail: Exploring Psychological Jailbreaks via Multi-Turn Persuasion of LLM Policies

Large language models (LLMs) are increasingly deployed in education, healthcare, policy advising, and other interactive settings, where users engage them as sustained social interlocutors rather than one-shot query engines. This shift makes jailbreaks a growing safety threat, yet most research emphasizes single-turn prompt optimization or iterative attack refinement, leaving psychologically grounded multi-turn vulnerabilities underexplored. We present PsychJail, a psychology-guided framework for red teaming aligned LLMs through theory-grounded, multi-turn persuasion. PsychJail maps established social-psychological persuasion techniques into a tactic-conditioned attack policy. It factorizes each attacker action into a Change-of-Meaning analysis, tactic selection, and victim-visible message, operationalizing the Persuasion Knowledge Model (PKM). The policy is refined with trajectory-level reinforcement learning using a PKM-gated reward that credits early jailbreak success only when every turn contains a well-formed Change-of-Meaning analysis. Across four aligned victim models, PsychJail achieves the highest average attack success rate (87.3%) and outperforms strong single-turn and multi-turn baselines on every model. We also measure susceptibility at the action that breaks each victim, revealing four distinct model-level fingerprints that identify which persuasion levers affect each model and how broadly. These fingerprints help explain cross-model transfer asymmetry. We interpret them as four candidate psychological profiles-rationalist, credibility-driven, narrative-monoculture, and broadly persuadable-while treating this interpretation as a conjecture requiring future validation. Our findings establish psychological jailbreaks as a distinct red-teaming frontier for increasingly interactive LLMs.

cs.AI↗

arXiv2026

Cyclic-Prefix-Free OFDM With Tail-Reuse Reconstruction for Distributed Acoustic Sensing

Orthogonal frequency-division multiplexing (OFDM) enables frequency-domain reconstruction of the distributed Rayleigh backscatter channel in coherent distributed acoustic sensing (DAS), but an explicitly transmitted cyclic prefix (CP) lengthens the probing period and reduces the slow-time Nyquist limit of each range cell. We investigate a repeated cyclic-prefix-free OFDM waveform for DAS, in which the tail of the preceding useful block serves as a virtual cyclic extension. A finite-memory range condition for tail-reuse reconstruction is derived, and circular folding is identified when the useful period is shorter than the channel memory. For fixed useful-block length and fiber memory, removing the explicit CP increases the period-limited highest unaliased vibration frequency without changing the occupied-bandwidth-limited spatial resolution. In a 5.2-km numerical configuration, a 75-MSa/s processing rate and a 4096-sample useful block give a 54.61-us probing period and a 9.16-kHz slow-time Nyquist limit. Simulations verify tail reuse, the predicted folding boundary, and recovery of 100 vibration events from 800 Hz to 8800 Hz. Bandwidth-scaling simulations further show that joint processing of fine spatial observations improves differential-phase reliability and reconstruction SNR at a fixed reporting interval. Experiments on a 5.2-km coherent DAS link with 111.984-MHz occupied OFDM bandwidth blindly localize 500-Hz and 3-kHz PZT-induced vibrations at 5063.7 m and 5070.1 m, respectively, and recover their waveforms and spectra. The results demonstrate feasible tail-reuse channel reconstruction and quantify the extension of the unaliased vibration bandwidth.

eess.SP↗

arXiv2026

Fourth-Order Cyclostationary Analysis of Power-Based Nonlinear Gardner Timing Error Detectors in Coherent Optical Systems

Power-based nonlinear Gardner timing error detectors (TEDs) can enhance clock-tone (CT) extraction in low-roll-off and bandwidth-limited coherent optical systems. However, their nonlinear power-domain operations make the extracted CT components depend on higher-order cyclic statistics, which cannot be fully characterized by second-order cyclostationary analysis. In this paper, we develop a fourth-order cyclostationary analytical framework for power-based Gardner-type TEDs, using the square-Gardner TED (SG-TED) as a representative case. We show that the SG-TED CT originates from the symbol-rate cyclic component of the power-process autocorrelation function (CAF), revealing its fourth-order cyclic-statistical origin in the received complex field. Through moment-cumulant decomposition, the CT component is separated into a Wick-reducible term and a cumulant-related non-Gaussian term, which respectively explain its connection to the conventional Gardner/Godard mechanism and its modulation- and distribution-dependent behavior. The framework further characterizes the effects of pulse shaping, probabilistic shaping, polarization rotation, and polarization-mode dispersion (PMD), revealing CT-response characteristics fundamentally different from second-order TEDs. Numerical evaluations and waveform-level Monte Carlo simulations validate the analysis and demonstrate the framework as a unified statistical basis for SG-TED and related power-based Gardner-type TEDs.

eess.SP↗

arXiv2026

Causal Scaffolding for Physical Reasoning: A Benchmark for Causally-Informed Physical World Understanding in VLMs

Understanding and reasoning about the physical world is the foundation of intelligent behavior, yet state-of-the-art vision-language models (VLMs) still fail at causal physical reasoning, often producing plausible but incorrect answers. To address this gap, we introduce CausalPhys, a benchmark of over 3,000 carefully curated video- and image-based questions spanning four domains: Perception, Anticipation, Intervention, and Goal Orientation. Each question is paired with an expert-annotated causal graph capturing object-attribute-event dependencies, enabling interpretable and fine-grained evaluation of causal understanding. Building on this, we formulate a causal-graph-grounded metric that quantitatively measures how well a model's chain-of-thought reasoning aligns with the correct causal relations, moving beyond answer-only accuracy and enabling systematic diagnosis of VLMs' causal reasoning failures. Using this metric, we conduct a comprehensive analysis of leading VLMs, revealing systematic gaps in capturing causal dependencies and underscoring the need for causality-aware learning. To address these limitations, we further propose Causal Rationale-informed Fine-Tuning (CRFT), which explicitly aligns VLM reasoning with causal structures. Extensive experiments demonstrate that CRFT substantially enhances both reasoning accuracy and interpretability across multiple model backbones. By unifying dataset curation, causal evaluation, and causality-informed learning, CausalPhys establishes a strong foundation for advancing modern VLMs toward causally grounded physical reasoning.

cs.DB↗

arXiv2025

PAN: A World Model for General, Actionable, and Long-Horizon World Simulation

A world model is a cognitive simulator of the real-world environment allowing biological agents to reason about how the world evolves, whether spontaneously or in response to their actions, and accordingly to plan and strategize. In building Artificial Intelligence (AI) systems, world models represent the next frontier beyond large language models (LLMs) to enable physical and embodied intelligence in AI agents, allowing them to perform decision-making through simulative reasoning and reinforcement-learning through simulative trials. Recent advancements in world modeling have yielded impressive progress in video generation, 3-D scene evolution, robotic dynamics, and game simulation, but limitations persist in general, open-domain, action-driven prediction, long-horizon consistency, and abstract reasoning and planning. Moreover, fundamental architectural questions, whether it be state representation, information flow, or training objectives, remain unresolved. In this paper, we introduce PAN, a world model built on the Generative Latent Prediction (GLP) architecture. GLP combines stateful latent representations of world states; an encoder--decoder closed-loop information flow; an LLM/diffusion-based mixed reasoning backbone; and a non-degenerate generative reconstruction objective whose fidelity is ``dampable'' to balance fine-grained detail against semantic saliency. Compared to several existing systems, PAN demonstrates advantages beyond standard video generation in action-conditioned world simulation, long-horizon forecasting, and simulative reasoning and planning, capabilities we argue should serve as the primary criteria for evaluating world models.

cs.CV↗

arXiv2025

Envelope Control Enabled Probabilistic Shaping for Peak Power Constrained IM DD Systems

Probabilistic shaping (PS) has attracted significant attention in intensity-modulation and direct-detection (IM-DD) systems. However, due to the unique system model and inherent constraints, the effective application of the PS technique is still an open question in IM-DD systems, particularly in systems with memory effects. In this paper, a novel indirect PS scheme tailored for peak power constrained (PPC) IM-DD systems is proposed. The key idea lies in strategically controlling the signal envelope to mitigate memory-induced impairments, such as nonlinearity, overshoot, peak-to-average power ratio enhancement, etc. The proposed scheme incorporates a dynamic selective mapping (DSLM) mechanism at the transmitter, enabling an untypical bit-to-symbol mapping in which the current symbol is not only determined by the current bits pattern but also by previously generated symbols within a specified memory length. At the receiver side, a turbo equalizer with a modified M-BCJR algorithm is proposed to achieve the recovery of ambiguous bits induced by DSLM. Experimental verification in a 56GBaud PAM8 system demonstrates that the proposed scheme exhibits 1dB receiver sensitivity improvement over 2km single-mode fiber transmission. In addition, the proposed scheme has also been demonstrated to be compatible with the typical probabilistic amplitude shaping architecture, enabling a simple and fine-granularity rate adaptation capability. To the best of our knowledge, this work opens a new sight for the application of the PS technique in PPC IM-DD systems with memory effects.

eess.SP↗

arXiv2024

Diffusion Meets Options: Hierarchical Generative Skill Composition for Temporally-Extended Tasks

Safe and successful deployment of robots requires not only the ability to generate complex plans but also the capacity to frequently replan and correct execution errors. This paper addresses the challenge of long-horizon trajectory planning under temporally extended objectives in a receding horizon manner. To this end, we propose DOPPLER, a data-driven hierarchical framework that generates and updates plans based on instruction specified by linear temporal logic (LTL). Our method decomposes temporal tasks into chain of options with hierarchical reinforcement learning from offline non-expert datasets. It leverages diffusion models to generate options with low-level actions. We devise a determinantal-guided posterior sampling technique during batch generation, which improves the speed and diversity of diffusion generated options, leading to more efficient querying. Experiments on robot navigation and manipulation tasks demonstrate that DOPPLER can generate sequences of trajectories that progressively satisfy the specified formulae for obstacle avoidance and sequential visitation. Demonstration videos are available online at: https://philiptheother.github.io/doppler/.

cs.RO↗

arXiv2024

Pandora: Towards General World Model with Natural Language Actions and Video States

World models simulate future states of the world in response to different actions. They facilitate interactive content creation and provides a foundation for grounded, long-horizon reasoning. Current foundation models do not fully meet the capabilities of general world models: large language models (LLMs) are constrained by their reliance on language modality and their limited understanding of the physical world, while video models lack interactive action control over the world simulations. This paper makes a step towards building a general world model by introducing Pandora, a hybrid autoregressive-diffusion model that simulates world states by generating videos and allows real-time control with free-text actions. Pandora achieves domain generality, video consistency, and controllability through large-scale pretraining and instruction tuning. Crucially, Pandora bypasses the cost of training-from-scratch by integrating a pretrained LLM (7B) and a pretrained video model, requiring only additional lightweight finetuning. We illustrate extensive outputs by Pandora across diverse domains (indoor/outdoor, natural/urban, human/robot, 2D/3D, etc.). The results indicate great potential of building stronger general world models with larger-scale training.

cs.CV↗

arXiv2024

Unified Generation, Reconstruction, and Representation: Generalized Diffusion with Adaptive Latent Encoding-Decoding

The vast applications of deep generative models are anchored in three core capabilities -- generating new instances, reconstructing inputs, and learning compact representations -- across various data types, such as discrete text/protein sequences and continuous images. Existing model families, like variational autoencoders (VAEs), generative adversarial networks (GANs), autoregressive models, and (latent) diffusion models, generally excel in specific capabilities and data types but fall short in others. We introduce Generalized Encoding-Decoding Diffusion Probabilistic Models (EDDPMs) which integrate the core capabilities for broad applicability and enhanced performance. EDDPMs generalize the Gaussian noising-denoising in standard diffusion by introducing parameterized encoding-decoding. Crucially, EDDPMs are compatible with the well-established diffusion model objective and training recipes, allowing effective learning of the encoder-decoder parameters jointly with diffusion. By choosing appropriate encoder/decoder (e.g., large language models), EDDPMs naturally apply to different data types. Extensive experiments on text, proteins, and images demonstrate the flexibility to handle diverse data and tasks and the strong improvement over various existing models.

cs.LG↗

arXiv2024

LTLDoG: Satisfying Temporally-Extended Symbolic Constraints for Safe Diffusion-based Planning

Operating effectively in complex environments while complying with specified constraints is crucial for the safe and successful deployment of robots that interact with and operate around people. In this work, we focus on generating long-horizon trajectories that adhere to novel static and temporally-extended constraints/instructions at test time. We propose a data-driven diffusion-based framework, LTLDoG, that modifies the inference steps of the reverse process given an instruction specified using finite linear temporal logic ($\text{LTL}_f$). LTLDoG leverages a satisfaction value function on $\text{LTL}_f$ and guides the sampling steps using its gradient field. This value function can also be trained to generalize to new instructions not observed during training, enabling flexible test-time adaptability. Experiments in robot navigation and manipulation illustrate that the method is able to generate trajectories that satisfy formulae that specify obstacle avoidance and visitation sequences. Code and supplementary material are available online at https://github.com/clear-nus/ltldog.

cs.RO↗

arXiv2023

Composable Text Controls in Latent Space with ODEs

Real-world text applications often involve composing a wide range of text control operations, such as editing the text w.r.t. an attribute, manipulating keywords and structure, and generating new text of desired properties. Prior work typically learns/finetunes a language model (LM) to perform individual or specific subsets of operations. Recent research has studied combining operations in a plug-and-play manner, often with costly search or optimization in the complex sequence space. This paper proposes a new efficient approach for composable text operations in the compact latent space of text. The low-dimensionality and differentiability of the text latent vector allow us to develop an efficient sampler based on ordinary differential equations (ODEs) given arbitrary plug-in operators (e.g., attribute classifiers). By connecting pretrained LMs (e.g., GPT2) to the latent space through efficient adaption, we then decode the sampled vectors into desired text sequences. The flexible approach permits diverse control operators (sentiment, tense, formality, keywords, etc.) acquired using any relevant data from different domains. Experiments show that composing those operators within our approach manages to generate or edit high-quality text, substantially improving over previous methods in terms of generation quality and efficiency.

cs.CL↗

arXiv2023

Synslator: An Interactive Machine Translation Tool with Online Learning

Interactive machine translation (IMT) has emerged as a progression of the computer-aided translation paradigm, where the machine translation system and the human translator collaborate to produce high-quality translations. This paper introduces Synslator, a user-friendly computer-aided translation (CAT) tool that not only supports IMT, but is adept at online learning with real-time translation memories. To accommodate various deployment environments for CAT services, Synslator integrates two different neural translation models to handle translation memories for online learning. Additionally, the system employs a language model to enhance the fluency of translations in an interactive mode. In evaluation, we have confirmed the effectiveness of online learning through the translation models, and have observed a 13% increase in post-editing efficiency with the interactive functionalities of Synslator. A tutorial video is available at:https://youtu.be/K0vRsb2lTt8.

cs.CL↗

arXiv2022

Safety-Constrained Policy Transfer with Successor Features

In this work, we focus on the problem of safe policy transfer in reinforcement learning: we seek to leverage existing policies when learning a new task with specified constraints. This problem is important for safety-critical applications where interactions are costly and unconstrained policies can lead to undesirable or dangerous outcomes, e.g., with physical robots that interact with humans. We propose a Constrained Markov Decision Process (CMDP) formulation that simultaneously enables the transfer of policies and adherence to safety constraints. Our formulation cleanly separates task goals from safety considerations and permits the specification of a wide variety of constraints. Our approach relies on a novel extension of generalized policy improvement to constrained settings via a Lagrangian formulation. We devise a dual optimization algorithm that estimates the optimal dual variable of a target task, thus enabling safe transfer of policies derived from successor features learned on source tasks. Our experiments in simulated domains show that our approach is effective; it visits unsafe states less frequently and outperforms alternative state-of-the-art methods when taking safety constraints into account.

cs.LG↗

arXiv2020

Open-Set Hypothesis Transfer with Semantic Consistency

Unsupervised open-set domain adaptation (UODA) is a realistic problem where unlabeled target data contain unknown classes. Prior methods rely on the coexistence of both source and target domain data to perform domain alignment, which greatly limits their applications when source domain data are restricted due to privacy concerns. This paper addresses the challenging hypothesis transfer setting for UODA, where data from source domain are no longer available during adaptation on target domain. We introduce a method that focuses on the semantic consistency under transformation of target data, which is rarely appreciated by previous domain adaptation methods. Specifically, our model first discovers confident predictions and performs classification with pseudo-labels. Then we enforce the model to output consistent and definite predictions on semantically similar inputs. As a result, unlabeled data can be classified into discriminative classes coincided with either source classes or unknown classes. Experimental results show that our model outperforms state-of-the-art methods on UODA benchmarks.

cs.CV↗