arXiv subjects

Zhizhen Qin

Publications and source records attributed to Zhizhen Qin.

13 recordsLinked to original sources

arXiv2026

Towards Mitigating Excessive Forgetting in LLM Unlearning via Entanglement-Guidance with Proxy Constraint

Large language models (LLMs) are trained on massive datasets that may include private or copyrighted content. Due to growing privacy and ownership concerns, data owners may request the removal of their data from trained models. Machine unlearning provides a practical solution by removing the influence of specific data without full retraining. However, most existing methods still suffer from over-unlearning due to the lack of a principled mechanism to regulate the forgetting boundary, leading to unnecessary utility degradation and heightened privacy and robustness risks. In this work, we propose EGUP (Entanglement-Guided Unlearning with Proxy Constraint), a novel framework that leverages entanglement and proxy constraint to guide the unlearning process while mitigating over-unlearning. Within each iteration, EGUP employs inter-sample entanglement to adaptively reweight the unlearning strength, assigning greater unlearning efforts to forget samples that are semantically closer to retained knowledge. Across iterations, EGUP leverages intra-sample entanglement to track the representation shift of each forget sample and dynamically adjust its unlearning effort. In addition, we incorporate a proxy constraint that approximates the model's expected outputs after unlearning, forming a reference boundary that softly regularizes the unlearning process. EGUP is compatible with existing gradient-based objectives and serves as a plug-and-play enhancement. We evaluate EGUP on the TOFU and MUSE benchmarks, demonstrating consistent improvements in the unlearning-utility trade-off across multiple LLMs. Moreover, EGUP achieves performance close to the retrained model while remaining scalable and robust.

cs.LG↗

arXiv2026

MIRAGE: Misleading Retrieval-Augmented Generation via Black-box and Query-agnostic Poisoning Attacks

Retrieval-Augmented Generation (RAG) systems enhance LLMs with external knowledge but introduce a critical attack surface: corpus poisoning. While recent studies have demonstrated the potential of such attacks, they typically rely on impractical assumptions, such as white-box access or known user queries, thereby underestimating the difficulty of real-world exploitation. In this paper, we bridge this gap by proposing MIRAGE, a novel multi-stage poisoning pipeline designed for strict black-box and query-agnostic environments. Operating on surrogate model feedback, MIRAGE functions as an automated optimization framework that integrates three key mechanisms: it utilizes persona-driven query synthesis to approximate latent user search distributions, employs semantic anchoring to imperceptibly embed these intents for high retrieval visibility, and leverages an adversarial variant of Test-Time Preference Optimization (TPO) to maximize persuasion. To rigorously evaluate this threat, we construct a new benchmark derived from three long-form, domain-specific datasets. Extensive experiments demonstrate that MIRAGE significantly outperforms existing baselines in both attack efficacy and stealthiness, exhibiting remarkable transferability across diverse retriever-LLM configurations and highlighting the urgent need for robust defense strategies.

cs.CR↗

arXiv2026

QuCtrl-BELL: A Compiler-Driven Sub-Microsecond Feedback Control Stack for Scalable Trapped-Ion Quantum Experiments

As trapped-ion quantum computing scales to larger qubit registers and more complex control protocols, classical control systems face a fundamental tradeoff: sub-microsecond board-level feedback requires tight hardware coupling, whereas maintainability and extensibility require clean, modular software abstractions. This paper presents QuCtrl-BELL (Bell), a compiler-driven software stack for trapped-ion quantum control. The design resolves this tradeoff by decoupling control flow -- including loops, branches, and synchronization -- from hardware state data. A Python-embedded domain-specific language (DSL) is lowered through a six-stage transpilation pipeline covering control flow graph (CFG) construction, static single-assignment (SSA) conversion, liveness analysis, and graph-coloring register allocation. The compiler generates deterministic distributed board-level programs and compact step-table data. A cross-board synchronization protocol supports feedback loops with latency below 700~ns without host intervention. Bell is deployed and evaluated on the QuCtrl-BELL platform (RISC-V + PXIe), demonstrating that a compiler-based infrastructure can provide programmability, deterministic timing, and modularity for scalable trapped-ion quantum control.

quant-ph↗

arXiv2026

Intent-aligned Formal Specification Synthesis via Traceable Refinement

Large language models are increasingly used to generate code from natural language, but ensuring correctness remains challenging. Formal verification offers a principled way to obtain such guarantees by proving that a program satisfies a formal specification. However, specifications are frequently missing in real-world codebases, and writing high-quality specifications remains expensive and expertise-intensive. We present VeriSpecGen, a traceable refinement framework that synthesizes intent-aligned specifications in Lean through requirement-level attribution and localized repair. VeriSpecGen decomposes natural language into atomic requirements and generates requirement-targeted tests with explicit traceability maps to validate generated specifications. When validation fails, traceability maps attribute failures to specific requirements, enabling targeted clause-level repairs. VeriSpecGen achieve 86.6% on VERINA SpecGen task using Claude Opus 4.5, improving over baselines by up to 31.8 points across different model families and scales. Beyond inference-time gains, we generate 343K training examples from VeriSpecGen refinement trajectories and demonstrate that training on these trajectories substantially improves specification synthesis by 62-106% relative and transfers gains to general reasoning abilities.

cs.LG↗

arXiv2025

The polar debris disc around 99 Herculis: A potential signpost for polar circumbinary planets

The nearby binary star system 99 Herculis (99 Her) is host to the only known polar-aligned circumbinary debris disc. We investigate the hypothesis that the narrow structure of this circumbinary disc is sculpted by the gravitational influence of one or more unseen polar circumbinary planets. We first establish the theoretically viable parameter space for a sculpting planet by considering dynamical stability and clearing mechanisms, including the chaotic zone, Hill radius, diffusion, and polar alignment timescales. We then use $N$-body simulations to test three specific architectures: a single planet interior to the disc, a single planet exterior, and a two-planet system bracketing the disc. Our simulations demonstrate that single-planet models are insufficient to reproduce the observed morphology, as they can only truncate one edge of the disc while leaving the other dynamically extended. In contrast, the two-planet shepherding model successfully carves both the inner and outer edges, confining the debris into a narrow, stable polar ring consistent with observations. We conclude that the structure of the 99 Her debris disc is most plausibly explained by the presence of two shepherding, polar circumbinary planets. We present a specific, testable model for this unique system, which elucidates the pivotal role of planetary bodies in sculpting the architecture of debris discs.

astro-ph.EP↗

arXiv2025

Breaking the Mass Inclination Degeneracy of Radial Velocity Measurements via Monitoring von Zeipel-Lidov-Kozai Cycles: Implications in the HD 41004 System

We investigate the dynamical stability of the S-type planet in the compact binary HD 41004. Using $N$-body simulations, we find that the planet could be dynamically stable at a mutual angle inclination up to $\sim75^\circ$. The von Zeipel-Lidov-Kozai (vZLK) mechanism becomes active when the mutual inclination is greater than 39.2$^\circ$. High-inclination orbits exhibit coupled oscillations in eccentricity and inclination, along with apsidal precession. Synthetic radial velocity (RV) modeling shows that these secular variations produce measurable signatures across a broad range of timescales, from full vZLK cycles to observationally accessible decades. For instance, a high mutual inclination at 75$^\circ$ can induce RV drifts exceeding 5 m s$^{-1}$ per planetary orbit ($\sim 1.9 \,\text{m s}^{-1}\, \text{yr}^{-1}$) in circular binary configurations. The presence of eccentric vZLK further accelerates these drifts, enhancing the detectability. Long-term RV observations of this system offer a unique pathway to dynamically constrain the orbital inclination, and thus determine the true mass of HD 41004 Ab. The degeneracy of mass-inclination is well known when using RV measurements alone. Our results highlight that HD 41004Ab and potentially other S-type planets in compact binaries are promising targets for breaking such a degeneracy by studying the dynamics induced by the vZLK mechanism through long-term high-precision RV monitoring.

astro-ph.EP↗

arXiv2025

Estimating Control Barriers from Offline Data

Learning-based methods for constructing control barrier functions (CBFs) are gaining popularity for ensuring safe robot control. A major limitation of existing methods is their reliance on extensive sampling over the state space or online system interaction in simulation. In this work we propose a novel framework for learning neural CBFs through a fixed, sparsely-labeled dataset collected prior to training. Our approach introduces new annotation techniques based on out-of-distribution analysis, enabling efficient knowledge propagation from the limited labeled data to the unlabeled data. We also eliminate the dependency on a high-performance expert controller, and allow multiple sub-optimal policies or even manual control during data collection. We evaluate the proposed method on real-world platforms. With limited amount of offline data, it achieves state-of-the-art performance for dynamic obstacle avoidance, demonstrating statistically safer and less conservative maneuvers compared to existing methods.

eess.SY↗

arXiv2024

SEEV: Synthesis with Efficient Exact Verification for ReLU Neural Barrier Functions

Neural Control Barrier Functions (NCBFs) have shown significant promise in enforcing safety constraints on nonlinear autonomous systems. State-of-the-art exact approaches to verifying safety of NCBF-based controllers exploit the piecewise-linear structure of ReLU neural networks, however, such approaches still rely on enumerating all of the activation regions of the network near the safety boundary, thus incurring high computation cost. In this paper, we propose a framework for Synthesis with Efficient Exact Verification (SEEV). Our framework consists of two components, namely (i) an NCBF synthesis algorithm that introduces a novel regularizer to reduce the number of activation regions at the safety boundary, and (ii) a verification algorithm that exploits tight over-approximations of the safety conditions to reduce the cost of verifying each piecewise-linear segment. Our simulations show that SEEV significantly improves verification efficiency while maintaining the CBF quality across various benchmark systems and neural network structures. Our code is available at https://github.com/HongchaoZhang-HZ/SEEV.

eess.SY↗

arXiv2024

Patching Approximately Safe Value Functions Leveraging Local Hamilton-Jacobi Reachability Analysis

Safe value functions, such as control barrier functions, characterize a safe set and synthesize a safety filter, overriding unsafe actions, for a dynamic system. While function approximators like neural networks can synthesize approximately safe value functions, they typically lack formal guarantees. In this paper, we propose a local dynamic programming-based approach to "patch" approximately safe value functions to obtain a safe value function. This algorithm, HJ-Patch, produces a novel value function that provides formal safety guarantees, yet retains the global structure of the initial value function. HJ-Patch modifies an approximately safe value function at states that are both (i) near the safety boundary and (ii) may violate safety. We iteratively update both this set of "active" states and the value function until convergence. This approach bridges the gap between value function approximation methods and formal safety through Hamilton-Jacobi (HJ) reachability, offering a framework for integrating various safety methods. We provide simulation results on analytic and learned examples, demonstrating HJ-Patch reduces the computational complexity by 2 orders of magnitude with respect to standard HJ reachability. Additionally, we demonstrate the perils of using approximately safe value functions directly and showcase improved safety using HJ-Patch.

cs.RO↗

arXiv2024

Sample-and-Bound for Non-Convex Optimization

Standard approaches for global optimization of non-convex functions, such as branch-and-bound, maintain partition trees to systematically prune the domain. The tree size grows exponentially in the number of dimensions. We propose new sampling-based methods for non-convex optimization that adapts Monte Carlo Tree Search (MCTS) to improve efficiency. Instead of the standard use of visitation count in Upper Confidence Bounds, we utilize numerical overapproximations of the objective as an uncertainty metric, and also take into account of sampled estimates of first-order and second-order information. The Monte Carlo tree in our approach avoids the usual fixed combinatorial patterns in growing the tree, and aggressively zooms into the promising regions, while still balancing exploration and exploitation. We evaluate the proposed algorithms on high-dimensional non-convex optimization benchmarks against competitive baselines and analyze the effects of the hyper parameters.

cs.AI↗

arXiv2023

Development of a SciFi-based beam monitor for COMET

COMET is a leading experiment to search for coherent conversion of $μ^- \mathrm{N}\to e^- \mathrm{N}$ with a high-intensity pulsed muon beamline, produced by the innovative slow extraction techniques. Therefore, it is critical to measure the characteristics of the muon beam. We set up a Muon Beam Monitor (MBM), where scintillation fibers (SciFi) weaved in the cross shape are coupled to silicon photomultipliers (SiPM), to measure the spatial profile and timing structure of the extracted muon beam for COMET. The MBM detector has been tested successfully with a proton beamline in China Spallation Neutron Source (CSNS) and taken data with good performance in the commissioning run called COMET Phase-$α$. Experience of the MBM development, such as the mechanical structure and electronics readout, and its beam measurement results will be shared.

physics.ins-det↗

arXiv2022

Policy Optimization with Advantage Regularization for Long-Term Fairness in Decision Systems

Long-term fairness is an important factor of consideration in designing and deploying learning-based decision systems in high-stake decision-making contexts. Recent work has proposed the use of Markov Decision Processes (MDPs) to formulate decision-making with long-term fairness requirements in dynamically changing environments, and demonstrated major challenges in directly deploying heuristic and rule-based policies that worked well in static environments. We show that policy optimization methods from deep reinforcement learning can be used to find strictly better decision policies that can often achieve both higher overall utility and less violation of the fairness requirements, compared to previously-known strategies. In particular, we propose new methods for imposing fairness requirements in policy optimization by regularizing the advantage evaluation of different actions. Our proposed methods make it easy to impose fairness constraints without reward engineering or sacrificing training efficiency. We perform detailed analyses in three established case studies, including attention allocation in incident monitoring, bank loan approval, and vaccine distribution in population networks.

cs.LG↗

arXiv2022

Quantifying Safety of Learning-based Self-Driving Control Using Almost-Barrier Functions

Path-tracking control of self-driving vehicles can benefit from deep learning for tackling longstanding challenges such as nonlinearity and uncertainty. However, deep neural controllers lack safety guarantees, restricting their practical use. We propose a new approach of learning almost-barrier functions, which approximately characterizes the forward invariant set for the system under neural controllers, to quantitatively analyze the safety of deep neural controllers for path-tracking. We design sampling-based learning procedures for constructing candidate neural barrier functions, and certification procedures that utilize robustness analysis for neural networks to identify regions where the barrier conditions are fully satisfied. We use an adversarial training loop between learning and certification to optimize the almost-barrier functions. The learned barrier can also be used to construct online safety monitors through reachability analysis. We demonstrate effectiveness of our methods in quantifying safety of neural controllers in various simulation environments, ranging from simple kinematic models to the TORCS simulator with high-fidelity vehicle dynamics simulation.

cs.RO↗