arXiv subjects

Zhenzhen Liu

Publications and source records attributed to Zhenzhen Liu.

17 recordsLinked to original sources

arXiv2026

STAIL: Semantic Text-Anchored Incremental Learning for Medical Imaging via Large Language Models

Deep learning models applied to medical image analysis suffer from severe catastrophic forgetting when continually adapting to new clinical tasks in dynamic environments. Mainstream incremental learning methods typically mitigate this by rehearsing raw historical images. However, this pixel-level rehearsal incurs significant storage overhead, raises privacy concerns, and fails to adequately capture the true data distribution with sparse exemplars. Inspired by human cognitive mechanisms, we propose a novel framework termed Semantic Text-Anchored Incremental Learning (STAIL) for sequential clinical tasks. To overcome the rehearsal bottleneck, STAIL introduces an asymmetric semantic consolidation buffer (SCB). By incorporating a minimal set of image anchors and extensive textual descriptions, the SCB enables dense semantic reconstruction of old tasks at a minimal storage cost. Furthermore, we design an LLM-derived Semantic Anchoring Mechanism (LSAM) that leverages the stable semantic space of frozen large language models as developmental priors. This mechanism explicitly anchors evolving visual features to textual representations, guiding and constraining plasticity and stability at both macroscopic and microscopic levels. Extensive experiments across three heterogeneous medical datasets, covering fundus, ultrasound, and X-ray imaging, demonstrate that STAIL acts as a highly effective plug-and-play module. It comprehensively enhances the performance of various existing baselines, achieving average gains of 2.24\% in AAA-AUC for sustained performance and 3.55\% in BWT-AUC for reduced forgetting. Code is available.

cs.CV↗

arXiv2026

Music Transcription with (Almost) No Supervision

Competitive music transcription models require large amounts of paired audio-score data, which is scarce due to collection costs, alignment difficulty, and copyright restrictions. Meanwhile, vast quantities of unpaired audio recordings and symbolic scores are freely available but have gone unused. We adopt a cycle-consistent translation framework in which a small amount of paired data acts as a minimal anchor, unlocking the full potential of the unpaired pool. We find that: unpaired data yields surprisingly large gains, especially under limited supervision; unpaired audio contributes more than unpaired scores; incorporating unlabeled audio from a new instrument during training improves transcription for that instrument without any paired supervision. Together, these results suggest that scaling unpaired data offers a practical path toward high-quality transcription for instruments where labeled data remains scarce.

cs.SD↗

arXiv2026

Crosstalk-free Chiral Anomaly Bulk States in Photonic Crystals

Ultracompact cladding-free waveguide arrays with zero inter-channel spacing and negligible crosstalk open a new avenue for high-density integrated photonic circuits. However, existing cladding-free waveguide arrays typically rely on conventional trivial bulk modes, making them highly susceptible to scattering losses at sharp bends or in the presence of obstacles and defects. To overcome this limitation, we theoretically propose and experimentally demonstrate a robust, crosstalk-free, and cladding-free photonic waveguide array based on chiral anomaly bulk states (CABSs) in photonic crystals. By interfacing distinct Dirac photonic crystals that host Dirac cones at different high-symmetry points ({\Gamma} and K) in the Brillouin zone and carefully engineering the boundary conditions, the boundary-induced CABSs in adjacent channels become effectively decoupled due to a large momentum separation, thereby eliminating inter-channel crosstalk. More importantly, we experimentally demonstrate that these crosstalk-free CABSs are robust to perturbations, including metallic obstacles, air defects, and sharp bends. We further extend the CABS-based waveguide array to two dimensions and demonstrate a cladding-free triangular resonator and a crosstalk-free waveguide crossing, both of which are previously unattainable. Our work establishes a new design paradigm for cladding-free, crosstalk-free, and ultracompact topological photonic devices, paving the way for robust, highly integrated photonic circuits.

physics.optics↗

arXiv2026

Attention in Space: Functional Roles of VLM Heads for Spatial Reasoning

Despite remarkable advances in large Vision-Language Models (VLMs), spatial reasoning remains a persistent challenge. In this work, we investigate how attention heads within VLMs contribute to spatial reasoning by analyzing their functional roles through a mechanistic interpretability lens. We introduce CogVSR, a dataset that decomposes complex spatial reasoning questions into step-by-step subquestions designed to simulate human-like reasoning via a chain-of-thought paradigm, with each subquestion linked to specific cognitive functions such as spatial perception or relational reasoning. Building on CogVSR, we develop a probing framework to identify and characterize attention heads specialized for these functions. Our analysis across diverse VLM families reveals that these functional heads are universally sparse, vary in number and distribution across functions. Notably, spatially specialized heads are fewer than those for other cognitive functions, highlighting their scarcity. We propose methods to activate latent spatial heads, improving spatial understanding. Intervention experiments further demonstrate their critical role in spatial reasoning: removing functional heads leads to performance degradation, while emphasizing them enhances accuracy. This study provides new interpretability driven insights into how VLMs attend to space and paves the way for enhancing complex spatial reasoning in multimodal models.

cs.AI↗

arXiv2026

When the City Teaches the Car: Label-Free 3D Perception from Infrastructure

Building robust 3D perception for self-driving still relies heavily on large-scale data collection and manual annotation, yet this paradigm becomes impractical as deployment expands across diverse cities and regions. Meanwhile, modern cities are increasingly instrumented with roadside units (RSUs), static sensors deployed along roads and at intersections to monitor traffic. This raises a natural question: can the city itself help train the vehicle? We propose infrastructure-taught, label-free 3D perception, a paradigm in which RSUs act as stationary, unsupervised teachers for ego vehicles. Leveraging their fixed viewpoints and repeated observations, RSUs learn local 3D detectors from unlabeled data and broadcast predictions to passing vehicles, which are aggregated as pseudo-label supervision for training a standalone ego detector. The resulting model requires no infrastructure or communication at test time. We instantiate this idea as a fully label-free three-stage pipeline and conduct a concept-and-feasibility study in a CARLA-based multi-agent environment. With CenterPoint, our pipeline achieves 82.3% AP for detecting vehicles, compared to a fully supervised ego upper bound of 94.4%. We further systematically analyze each stage, evaluate its scalability, and demonstrate complementarity with existing ego-centric label-free methods. Together, these results suggest that city infrastructure itself can potentially provide a scalable supervisory signal for autonomous vehicles, positioning infrastructure-taught learning as a promising orthogonal paradigm for reducing annotation cost in 3D perception.

cs.CV↗

arXiv2026

On the Feasibility and Opportunity of Autoregressive 3D Object Detection

LiDAR-based 3D object detectors typically rely on proposal heads with hand-crafted components like anchor assignment and non-maximum suppression (NMS), complicating training and limiting extensibility. We present AutoReg3D, an autoregressive 3D detector that casts detection as sequence generation. Given point-cloud features, AutoReg3D emits objects in a range-causal (near-to-far) order and encodes each object as a short, discrete-token sequence consisting of its center, size, orientation, velocity, and class. This near-to-far ordering mirrors LiDAR geometry--near objects occlude far ones but not vice versa--enabling straightforward teacher forcing during training and autoregressive decoding at test time. AutoReg3D is compatible across diverse point-cloud or backbones and attains competitive nuScenes performance without anchors or NMS. Beyond parity, the sequential formulation unlocks language-model advances for 3D perception, including GRPO-style reinforcement learning for task-aligned objectives. These results position autoregressive decoding as a viable, flexible alternative for LiDAR-based detection and open a path to importing modern sequence-modeling tools into 3D perception.

cs.CV↗

arXiv2025

Benchmark Datasets for Lead-Lag Forecasting on Social Platforms

Social and collaborative platforms emit multivariate time-series traces in which early interactions -- such as views, likes, or downloads -- are followed, sometimes months or years later, by higher impact like citations, sales, or reviews. We formalize this setting as Lead-Lag Forecasting (LLF): given an early usage channel (the lead), predict a correlated but temporally shifted outcome channel (the lag). Despite the ubiquity of such patterns, LLF has not been treated as a unified forecasting problem within the time-series community, largely due to the absence of standardised datasets. To anchor research in LLF, here we present two high-volume benchmark datasets: arXiv (accesses -> citations of 2.3M papers) and GitHub (pushes/stars -> forks of 3M repositories). Our datasets provide ideal testbeds for lead-lag forecasting, by capturing long-horizon dynamics across years, spanning the full spectrum of outcomes, and avoiding survivorship bias in sampling. We documented all technical details of data curation and cleaning, verified the presence of lead-lag dynamics through statistical and classification tests, and benchmarked parametric and non-parametric baselines for regression. Our study establishes LLF as a novel forecasting paradigm and lays an empirical foundation for its systematic exploration in social and usage data.

cs.LG↗

arXiv2025

Mixed Signals: A Diverse Point Cloud Dataset for Heterogeneous LiDAR V2X Collaboration

Vehicle-to-everything (V2X) collaborative perception has emerged as a promising solution to address the limitations of single-vehicle perception systems. However, existing V2X datasets are limited in scope, diversity, and quality. To address these gaps, we present Mixed Signals, a comprehensive V2X dataset featuring 45.1k point clouds and 240.6k bounding boxes collected from three connected autonomous vehicles (CAVs) equipped with two different configurations of LiDAR sensors, plus a roadside unit with dual LiDARs. Our dataset provides point clouds and bounding box annotations across 10 classes, ensuring reliable data for perception training. We provide detailed statistical analysis on the quality of our dataset and extensively benchmark existing V2X methods on it. The Mixed Signals dataset is ready-to-use, with precise alignment and consistent annotations across time and viewpoints. Dataset website is available at https://mixedsignalsdataset.cs.cornell.edu/.

cs.CV↗

arXiv2024

DiffuBox: Refining 3D Object Detection with Point Diffusion

Ensuring robust 3D object detection and localization is crucial for many applications in robotics and autonomous driving. Recent models, however, face difficulties in maintaining high performance when applied to domains with differing sensor setups or geographic locations, often resulting in poor localization accuracy due to domain shift. To overcome this challenge, we introduce a novel diffusion-based box refinement approach. This method employs a domain-agnostic diffusion model, conditioned on the LiDAR points surrounding a coarse bounding box, to simultaneously refine the box's location, size, and orientation. We evaluate this approach under various domain adaptation settings, and our results reveal significant improvements across different datasets, object classes and detectors. Our PyTorch implementation is available at \href{https://github.com/cxy1997/DiffuBox}{https://github.com/cxy1997/DiffuBox}.

cs.CV↗

arXiv2024

Correction with Backtracking Reduces Hallucination in Summarization

Abstractive summarization aims at generating natural language summaries of a source document that are succinct while preserving the important elements. Despite recent advances, neural text summarization models are known to be susceptible to hallucinating (or more correctly confabulating), that is to produce summaries with details that are not grounded in the source document. In this paper, we introduce a simple yet efficient technique, CoBa, to reduce hallucination in abstractive summarization. The approach is based on two steps: hallucination detection and mitigation. We show that the former can be achieved through measuring simple statistics about conditional word probabilities and distance to context words. Further, we demonstrate that straight-forward backtracking is surprisingly effective at mitigation. We thoroughly evaluate the proposed method with prior art on three benchmark datasets for text summarization. The results show that CoBa is effective and efficient in reducing hallucination, and offers great adaptability and flexibility. Code can be found at https://github.com/zhenzhel/CoBa.

cs.CL↗

arXiv2024

Hypermultiplexed off-chip hologram by on-chip integrated metasurface

The waveguide-integrated metasurface introduces a novel photonic chip capable of converting guided modes into free-space light. This enables functions such as off-chip beam focusing, steering, and imaging. The challenge lies in achieving hypermultiplexing across diverse parameters, including guided-wave mode type, direction, polarization, and notably, multiple wavelengths. Here, we introduce a comprehensive end-to-end inverse design framework, rooted in a physical model, for the multifunctional design of on-chip metasurfaces. This framework allows for metasurface optimization through a target-field-driven iteration process. We demonstrate a hypermultiplexed on-chip metasurface capable of generating red-green-blue holograms at multiple target planes, with both independent and cooperative control over guided-wave direction. Significantly, the proposed method streamlines the design process utilizing only the positions of meta-atoms as the design variable. We demonstrate 9 independent holographic channels through a combination of wavelength and distance multiplexing. Moreover, by incorporating the excitation direction into the design, the metasurface produces a total of 36 distinct holograms. The robustness of these results against fabrication discrepancies is validated through 3D full-wave electromagnetic simulations, aligning well with advanced manufacturing techniques. Our research presents a universal design framework for the development of multifunctional on-chip metasurfaces, opening up new avenues for a wide range of applications.

physics.optics↗

arXiv2024

Topological Corner Modes by Composite Wannier States in Glide-Symmetric Photonic Crystal

Second-order topological insulators can be characterized by their bulk polarization, which is believed to be intrinsically connected to the center of the Wannier function. In this study, we demonstrate the existence of second-order topological insulators that feature a pair of partially degenerate photonic bands. These arise from the nonsymmorphic glide symmetry in an all-dielectric photonic crystal. The center of the maximally localized Wannier function (MLWF) is consistently located at the origin but is not equivalent with respect to the sum of constituent polarizations. As a result, topological corner modes can be identified by the distinctly hybridized MLWFs that truncate at the sample boundary. Through full-wave numerical simulations paired with microwave experiments, the second-order topology is clearly confirmed and characterized. These topological corner states exhibit notably unique modal symmetries, which are made possible by the inversion of the Wannier bands. Our results provide an alternative approach to explore higher-order topological physics with significant potential for applications in integrated and quantum photonics.

cond-mat.mes-hall↗

arXiv2024

Detecting Out-of-Distribution Objects through Class-Conditioned Inpainting

Recent object detectors have achieved impressive accuracy in identifying objects seen during training. However, real-world deployment often introduces novel and unexpected objects, referred to as out-of-distribution (OOD) objects, posing significant challenges to model trustworthiness. Modern object detectors are typically overconfident, making it unreliable to use their predictions alone for OOD detection. To address this, we propose leveraging an auxiliary model as a complementary solution. Specifically, we utilize an off-the-shelf text-to-image generative model, such as Stable Diffusion, which is trained with objective functions distinct from those of discriminative object detectors. We hypothesize that this fundamental difference enables the detection of OOD objects by measuring inconsistencies between the models. Concretely, for a given detected object bounding box and its predicted in-distribution class label, we perform class-conditioned inpainting on the image with the object removed. If the object is OOD, the inpainted image is likely to deviate significantly from the original, making the reconstruction error a robust indicator of OOD status. Extensive experiments demonstrate that our approach consistently surpasses existing zero-shot and non-zero-shot OOD detection methods, establishing a robust framework for enhancing object detection systems in dynamic environments.

cs.LG↗

arXiv2023

Reward Finetuning for Faster and More Accurate Unsupervised Object Discovery

Recent advances in machine learning have shown that Reinforcement Learning from Human Feedback (RLHF) can improve machine learning models and align them with human preferences. Although very successful for Large Language Models (LLMs), these advancements have not had a comparable impact in research for autonomous vehicles -- where alignment with human expectations can be imperative. In this paper, we propose to adapt similar RL-based methods to unsupervised object discovery, i.e. learning to detect objects from LiDAR points without any training labels. Instead of labels, we use simple heuristics to mimic human feedback. More explicitly, we combine multiple heuristics into a simple reward function that positively correlates its score with bounding box accuracy, i.e., boxes containing objects are scored higher than those without. We start from the detector's own predictions to explore the space and reinforce boxes with high rewards through gradient updates. Empirically, we demonstrate that our approach is not only more accurate, but also orders of magnitudes faster to train compared to prior works on object discovery.

cs.CV↗

arXiv2023

Unsupervised Out-of-Distribution Detection with Diffusion Inpainting

Unsupervised out-of-distribution detection (OOD) seeks to identify out-of-domain data by learning only from unlabeled in-domain data. We present a novel approach for this task - Lift, Map, Detect (LMD) - that leverages recent advancement in diffusion models. Diffusion models are one type of generative models. At their core, they learn an iterative denoising process that gradually maps a noisy image closer to their training manifolds. LMD leverages this intuition for OOD detection. Specifically, LMD lifts an image off its original manifold by corrupting it, and maps it towards the in-domain manifold with a diffusion model. For an out-of-domain image, the mapped image would have a large distance away from its original manifold, and LMD would identify it as OOD accordingly. We show through extensive experiments that LMD achieves competitive performance across a broad variety of datasets. Code can be found at https://github.com/zhenzhel/lift_map_detect.

cs.CV↗

arXiv2022

Type-\textbf{III} corner state in second-order topological insulator by distinctly hybridized photonic Wannier functions

In the presence of crystalline symmetries, second-order topological insulators can be featured by the polarization which is believed identical to the Wannier center. In this Letter, we show that second-order topological insulators are present in the full process of topological phase transition between a pair of degenerate photonic bands resulting from the non-symmorphic glide symmetry in all-dielectric photonic crystals. Due to the coupling (hybridization) of the Wannier functions, the center of the maximally localized Wannier function always locates at the origin, not equal to the addition of constituent polarization. In this case, the corner states are ascribed to the hybridized Wannier functions obstructed by the boundary. In combination with the local density of states, the second-order topology is clearly confirmed and characterized. That ensures corner states which are fundamentally different to either the type-I corner state localized in particular sublattice, or the type-II corner state originated from interacting topological edge states. This type of corner states is regarded as type-III here and provides an alternative method to explore higher-order topological physics, which may lead to interesting applications in integrated and quantum photonics.

physics.optics↗

arXiv2015

Collective dark states controlled transmission in plasmonic slot waveguide with a stub coupled to a cavity dimer

We report collective dark states controlled transmission in metal-dielectric-metal waveguides with a stub coupled to two twin cavities, namely, plasmonic waveguide-stub-dimer systems. In absence of one individual cavity in the dimer, plasmon induced transparency (PIT) is possible when the cavity and the stub have the same resonance frequency. However, it is shown that the hybridized modes in the dimer collectively generate two dark states which make the stub-dimer "invisible" to the straight waveguide, splitting the original PIT peak into two in the transmission spectrum. Simultaneously, the original PIT peak becomes a dip due to dark state interaction, yielding anti-PIT-like modulation of the transmission. With full-wave electromagnetic simulation, we demonstrate that this transition is controlled by the dimer-stub separation and the dimer-stub relative position. All results are analytically described by the temporal coupled mode theory. Our results may be useful in designing densely integrated optical circuits, and in optical sensing and switching applications.

physics.optics↗