arXiv subjects

Li Zhang

Publications and source records attributed to Li Zhang.

At least 73 records · Page 4Linked to original sources

arXiv2026

Towards Robust Driving Perception: A Flexible Scale-Driven Family for Self-Supervised Monocular Depth Estimation

Self-Supervised Monocular Depth Estimation (MDE) has garnered attention in recent years due to its independence from ground truth. However, most existing models are limited to a single scale and exhibit considerable performance degradation in complex driving environments. Networks specifically designed to handle dynamic traffic participants tend to be overly complex, hindering their deployment on resource-constrained automotive edge devices. To address these limitations and move towards robust driving perception, we propose FlexDepth, a scale-driven and flexible family of self-supervised MDE models tailored for challenging road scenarios. FlexDepth employs a two-stage static-dynamic decoupled training strategy, enabling the independent assessment of confidence for both static backgrounds and dynamic road objects. Furthermore, it introduces a meticulously designed Scale-Driven Decoder (SDD) to dynamically select components based on scale size, facilitating efficient feature fusion and the output of high-precision depth maps. Extensive experiments on standard driving benchmarks demonstrate that without any auxiliary information, our model achieves state-of-the-art performance across arbitrary scales with minimal computational overhead. Our smallest model, Flex-Nano, requires only 0.7 GFLOPs and achieves 37.6 FPS on mobile platforms, ensuring reliable real-time perception while maintaining excellent zero-shot generalization. Our source code is available: https://github.com/startnew/flexdepth

cs.CV↗

arXiv2026

$\mathcal{F}$-Transitivity of Translation Semigroups on Directed Metric Trees

We study the $\mathcal F$-transitivity and topological $\mathcal F$-recurrence of left translation semigroups on weighted $L^p$-spaces over directed metric trees. Motivated by the recent work of Mangino and Vargas-Moreno on hypercyclicity and weak mixing for these semigroups, we investigate the different problem of $\mathcal F$-transitivity, where the entire return-time set is required to belong to a prescribed finitely invariant Furstenberg family. Assuming that the weight is $p$-admissible, we obtain necessary and sufficient integral conditions for both rooted and rootless trees, and establish the equivalence between $\mathcal F$-transitivity and topological $\mathcal F$-recurrence. In the rooted case, the criteria depend only on the weights along descendant edges. In the rootless case, an additional ancestor term and a measurable cancellation construction reflect the backward geometry of the tree. Our approach is based on measurable weighted minimization and quantitative estimates over subsets of the edge parameter interval. Examples on homogeneous rooted trees and rootless trees with a free left end show that branching may determine the dynamics, but forward branching alone does not guarantee $\mathcal F$-transitivity in the rootless setting.

math.FA↗

arXiv2026

Video Compression with Graph-inspired Neural Representation

Implicit Neural Representations (INR) provide a compact and content-adaptive paradigm for video compression, typically representing a video through shared network parameters and frame-indexed embeddings. Compared to conventional or autoencoder-based codecs, these approaches exploit temporal redundancy within videos in an implicit manner, which potentially results in sub-optimal compression performance. In this paper, we propose G-NeRV, a graph-inspired INR that explicitly improves temporal redundancy exploitation in the implicit latent space. Motivated by the total correlation principles in information theory, we construct a temporal neighborhood over frame embeddings and perform message passing to aggregate reusable information from neighboring frames through an adaptive gate controlling the injection of neighboring information. Inspired by the reference frame buffer in conventional video coding, a memory bank mechanism has been further designed to enable efficient temporal-neighbor retrieval under random frame-index sampling in INR training. This new representation model has been integrated into an advanced representation compression framework and compared with existing conventional and neural video codecs. The results show that the G-NeRV codec outperforms the state-of-the-art INR-based codec, NVRC, and the latest standard video codec, VVC VTM, by 8.86\% and 14.68\% (in BD-rate), respectively, measured by PSNR on the UVG dataset.

cs.CV↗

arXiv2026

Performance Optimization of Short Reach Optical Interconnects based on Direct Detection

Short-reach optical interconnects are evolving toward data rates beyond 400 Gb/s per lane, driven by the bandwidth and energy-efficiency requirements of AI-enabled datacenter networks. At these operating speeds, channel impairments, device nonlinearities, and hardware constraints limit the effectiveness of conventional transceiver design and digital signal processing (DSP). This paper presents a unified framework for the optimization of direct-detection optical interconnects, encompassing digital surrogate modeling, receiver-side DSP optimization, and end-to-end (E2E) transceiver learning. The proposed formulation provides a common perspective for model-based and machine-learning-based approaches, including linear and nonlinear equalization, lookup tables, decision trees, neural-network receivers, and E2E optimization. Their performance is discussed together with computational complexity and hardware implementation aspects, highlighting the associated trade-offs. We show, through simulations and experimental validations, that for a 40~Gb/s 10~km link, decision trees can outperform by 0.5--1~dB conventional linear equalization, with negligible hardware requirements. Moreover, we show that an E2E technique based on transformers can provide a gain up to 6~dB, highlighting the potential of joint transceiver optimization to improve the performance of next-generation short-reach optical links.

physics.optics↗

arXiv2026

Coronary Mask Guided Registration for Continuous Time 4D Cardiac CT Dataset Construction

Objective: Clinical cardiac CT multiphase reconstructions generally provide acceptable image quality in end-diastole (ED) or end-systole (ES) phases, but in other phases may exhibit motion artifacts, especially in the right coronary artery (RCA). This limits ground-truth availability in 4D cardiac CT imaging research. We aim to construct a 4D cardiac CT dataset that is generally suitable to serve as pseudo ground truth. Methods: We propose Coronary Mask Guided Registration (CMGR) to produce a motion-preserved, artifact-reduced, and continuous-time 4D cardiac CT sequence from the clinical multiphase reconstruction of each patient. For artifact reduction, CMGR uses the ED or ES phase as the reference phase and warps the reference volume with deformation fields to produce the sequence. For motion preservation, CMGR registers the reference phase to each non-reference phase of the multiphase reconstruction. To capture the motion of both the RCA and other cardiac structures in each registration, CMGR regularizes RCA masks and incorporates them into image-domain registration. Time-continuity is achieved by interpolating the deformation fields for non-reference phases to arbitrary times. Results: CMGR outperformed representative image-domain registration methods in capturing RCA motion and providing reasonable RCA shape, and showed competitive performance in capturing whole-heart motion. Additionally, CMGR reduced motion artifacts from clinical multiphase reconstructions, and intermediate CMGR frames generally provided plausible transitions between discrete cardiac phases. Conclusion: CMGR provides an effective approach for constructing continuous-time 4D cardiac CT datasets. Significance: The dataset can be used in system design simulations and in reconstruction algorithm development, thereby facilitating advances in cardiac CT imaging.

eess.IV↗

arXiv2026

Fine-Tuning of Transformer models with Frames

Parameter-Efficient Fine-Tuning (PEFT) strategies such as Low-Rank Adaptation (LoRA) are effective solutions for fine-tuning large-scale pre-trained models; however, their memory requirements scale with the size of the model, $\mathcal{O}(dr)$, where $d$ is the model's hidden dimension and $r$ is the rank. Our proposal, FrameFT, models the parameter update $ΔW$ with a sparse coefficient matrix in a Fusion Frame basis. Fusion Frames can be generated algorithmically and shared across model layers, enabling very efficient updates. Only the sparse coefficients of the basis expansion are stored/optimized, reducing the memory footprint. The sparse structure of the coefficient matrix in FrameFT and the sparsity in the Fusion Frames give large compute benefits, and our analysis provides formal convergence results. We evaluate the idea across a suite of supervised fine-tuning benchmarks, focusing on language tasks, but also report application to vision models. Our experiments show that FrameFT achieves performance on par with/exceeding state-of-the-art PEFT techniques, but needs far fewer trainable parameters.

cs.AI↗

arXiv2026

Energy-dependent extension of the Crab Nebula as a diagnostic of ultrahigh-energy hadronic emission

The energy-dependent apparent extension of the Crab Nebula, inconsistent with leptonic expectations between X-ray and TeV bands, can be naturally accounted for in a spatially resolved lepto-hadronic scenario. A transition to hadronic dominance occurs above $\sim400~\rm{TeV}$, with nearly pure hadronic emission at PeV energies. The predicted $68\%$ containment radius at $1.1~\rm{PeV}$ is $\sim1.4'$, about six times larger than in purely leptonic scenarios. This energy-dependent extension therefore provides an observable discriminator of the origin of ultra-high-energy gamma-ray emission in pulsar wind nebulae, offering a diagnostic of PeV hadronic acceleration in the Crab Nebula.

astro-ph.HE↗

arXiv2026

Maia 200: A Software Defined Dataflow System for Large-scale AI Acceleration

We introduce Maia 200, an advanced AI accelerator delivering high performance-10 145 Tflop/s FP4 and 5072 Tflop/s FP8 within a 750W TDP and 7 TB/s HBM bandwidth. Maia exemplifies a new class of Software Defined Locally Accessed Dataflow Architectures (SDLA), which explicitly program dataflow engines to orchestrate highly specialized memories and data movement engines. This approach shifts the focus from today's thread-centric to data-movement-centric architecture, improving efficiency and scalability. Our taxonomy of data management, inspired by Flynn's classification, highlights how SDLA addresses challenges in modern AI computing. Maia 200 achieves significant cost and energy savings while supporting massive parallelism for AI inference workloads, making it a compelling solution for next-generation high-performance computing systems.

cs.AR↗

arXiv2026

Visualizing flat-band spatial renormalization in rhombohedral graphene superlattices

Rhombohedral graphene/hBN moiré superlattices exhibit flat-band-driven emergent phases, including superconductivity and the fractional quantum anomalous Hall effect (FQAHE), yet the microscopic role of the moiré potential remains unclear. Here, using scanning tunneling microscopy, we visualize moiré-modulated spatial renormalization of flat bands in rhombohedral pentalayer and tetralayer graphene/hBN superlattices. We observe spatially hierarchical filling, manifested as periodic energy shifts of the flat bands at the moiré scale, leading to spatial reshaping of correlated states in the interacting regime. Remarkably, this modulation vanishes below a ~10 nm moiré period--the same threshold below which the FQAHE is absent. Theoretical modeling attributes this mechanism to atomic-corrugation-induced charge redistribution. Our work provides real-space visualization of moiré-engineered flat-band reconstruction, resolving a key link between moiré periodic potential and emergent topological order.

cond-mat.mes-hall↗

arXiv2026

crossMoDA Challenge: Evolution of Cross-Modality Domain Adaptation Techniques for Vestibular Schwannoma and Cochlea Segmentation from 2021 to 2023

The cross-Modality Domain Adaptation (crossMoDA) challenge series, initiated in 2021 in conjunction with the International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI), focuses on unsupervised cross-modality segmentation, learning from contrast-enhanced T1 (ceT1) and transferring to T2 MRI. The task is an extreme example of domain shift chosen to serve as a meaningful and illustrative benchmark. From a clinical application perspective, it aims to automate Vestibular Schwannoma (VS) and cochlea segmentation on T2 scans for more cost-effective VS management. Over time, the challenge objectives have evolved to enhance its clinical relevance. The challenge evolved from using single-institutional data and basic segmentation in 2021 to incorporating multi-institutional data and Koos grading in 2022, and by 2023, it included heterogeneous routine data and sub-segmentation of intra- and extra-meatal tumour components. In this work, we report the findings of the 2022 and 2023 editions and perform a retrospective analysis of the challenge progression over the years. The observations from the successive challenge contributions indicate that the number of outliers decreases with an expanding dataset. This is notable since the diversity of scanning protocols of the datasets concurrently increased. The winning approach of the 2023 edition reduced the number of outliers on the 2021 and 2022 testing data, demonstrating how increased data heterogeneity can enhance segmentation performance even on homogeneous data. However, the cochlea Dice score declined in 2023, likely due to the added complexity from tumour sub-annotations affecting overall segmentation performance. While progress is still needed for clinically acceptable VS segmentation, the plateauing performance suggests that a more challenging cross-modal task may better serve future benchmarking.

eess.IV↗

arXiv2026

Do Value Vectors in Deep Layers Need Context from the Residual Stream?

The success of the transformer architecture is in large part due to its use of attention layers. An attention layer follows the standard neural network paradigm: it takes the residual stream as input and thereby produces context-dependent query, key, and value vectors. However, we find that model performance meaningfully improves when deeper layers learn only a context-free value vector to preserve the original token information, without drawing on any context from the residual stream. When the model has access to this context-free value vector, adding back the context-dependent component provides little additional benefit for aggregate benchmark performance. Such context-free value vectors can be stored as sparse model parameters, eliminating the need to recompute or persistently cache these values. Through systematic ablations on the key design choices for such context-free value vectors, we propose Bank of Values (BoV), a new way of computing value vectors in attention by learning a lookup table of token-specific value vectors for each of the last third of layers. Across 135M and 780M models, BoV improves validation loss over standard attention and, at 780M, the average score across 21 benchmarks, matching the previous best method that adds token information to the value vector with less compute and memory.

cs.CL↗

arXiv2026

Asymmetric Capacity Allocation in Self-Refinement Pipelines

Self-refinement, typically structured as generation, critique, and revision, is a widely adopted paradigm for improving LLM generation and serves as a core mechanism in many LLM agents. While the three stages involve different cognitive demands, most existing approaches conveniently treat the model size as an implementation detail rather than a subject of study, which may lead to a waste of resources. Little work has systematically examined how model size affects each stage or whether effective self-refinement requires equally capable models for generation, critique, and revision. We present the first stage-wise model size study of the self-refinement pipeline on 5 benchmarks from different domains using 6 model sizes of Qwen3 and 4 model sizes of Gemma 3. We conclude that larger generators and refiners generally improve the pipeline, whereas an undersized refiner can even harm performance. Second, performance is highly insensitive to the size of the critic, although including even a small critic consistently outperforms omitting critique altogether. Our findings demonstrate that model capacity should not be allocated uniformly across self-refinement pipelines. Instead, different stages exhibit distinct size scaling characteristics, providing practical guidance for designing more computationally efficient multi-stage language model systems.

cs.LG↗

arXiv2026

First Observational Evidence for QPO-like Coevolution between Characteristic Damping Timescales and X-ray Time Lags among AGNs

Quasi-periodic oscillations (QPOs) and stochastic variability provide complementary probes of the inner accretion flow around supermassive black holes in active galactic nuclei (AGNs). Previous multi-epoch studies of narrow-line Seyfert 1 galaxy RE~J1034+396 revealed a coevolution between the QPO frequency and the X-ray time lag, but whether the stochastic variability participates in the same structural evolution has remained unclear. We analyze the multi-epoch XMM-Newton observations of RE~J1034+396 and model the soft (0.3--1~keV) and hard (1--4~keV) light curves separately using a damped random walk process. We obtain reliable characteristic damping timescales (CDTs) for 17 observations, with the soft-band CDT consistently longer than its hard-band counterpart. When combined with the X-ray time lag, the hard-band CDT traces a counterclockwise closed loop that closely resembles the previously reported QPO-frequency--time-lag loop, whereas the soft-band CDT exhibits more complex trajectory. Under plausible dynamical, thermal, and viscous interpretations, the hard-band CDT is associated with characteristic scales in the inner hot accretion flow/corona. The observed loops may represent different projections of a common cyclic evolution of the inner accretion flow/corona, as the three timing observables may respond to its dynamical, stochastic, and radiative properties on different timescales. These results provide the first observational support among AGNs for the physical picture in which changes in the spatial extent of the hot inner flow/corona simultaneously affect QPO and stochastic variability.

astro-ph.HE↗

arXiv2026

Signed list edge coloring in graphs of bounded treewidth

Vizing conjectured that the list edge chromatic number of any graph with maximum degree $Δ$ is at most $Δ+ 1$. This conjecture has been confirmed for several important classes of graphs, in particular, Lang proved that it holds for all graphs of treewidth $3$. In this paper, we introduce the list edge coloring of signed graphs, a framework that generalizes both classical list edge coloring and the signed edge coloring introduced by Behr. We extend Lang's result by proving the signed analogue of Vizing's conjecture for all signed graphs of treewidth $3$, as well as for signed graphs of treewidth $4$ with maximum degree $Δ\ge 10$.

math.CO↗

arXiv2026

A Reality Check of Language Models as Formalizers on Constraint Satisfaction Problems

Recent work shows superior performance when using large language models (LLMs) as formalizers instead of as end-to-end solvers for symbolic reasoning problems. Given the problem description, the LLM generates a formal program that derives a solution via an external solver. We systematically investigate the formalization capability of LLMs on real-life constraint satisfaction problems on 4 benchmarks, 6 LLMs, and 2 types of formal languages. We show that LLM-as-formalizer by no means trivializes the problem but underperforms LLM-as-solver in 15 out of 24 model-dataset combinations, despite the former's verifiability and interpretability. Although the formalization space is magnitudes smaller than the search space, our scaling analysis shows that LLM-as-formalizer still drastically degrades as problem complexity increases similar to LLM-as-solver. To better understand this limitation, we observe excessive, solver-like reasoning tokens that sometimes lead to hard-coded solutions, highlighting a key challenge for improving LLM-based formalization.

cs.CL↗

arXiv2026

BiCRVC: An Efficient Bidirectional Neural Video Compression Framework via Coupled Representation Coding

Neural video compression (NVC) has achieved strong compression performance, but practical random-access coding still faces two technical challenges: existing bidirectional NVCs (BVCs) usually require costly motion-first decoding, and reliable motion estimation is difficult under long-range bidirectional prediction. To address these issues, we present BiCRVC, an efficient bidirectional neural video compression framework based on coupled representation coding. Instead of coding motion and frame information with two separate codecs, BiCRVC transforms the motion representation and the current-frame latent into a unified latent representation for entropy coding. This design enables motion and frame information to be decoded from the same bitstream with one unified codec, while still reconstructing motion-aligned contexts for frame decoding. To improve motion accuracy, we introduce multi-candidate motion estimation (MCME), which combines multi-scale motion estimation and parallel accumulated motion estimation to better handle diverse and long-range motions. To reduce motion coding overhead, we further propose bidirectional motion feature propagation (BMFP), which reuses previously decoded motion features at both the encoder and decoder as temporal priors for conditional motion coding. In addition, coupled distortion training and random GOP structure training are used to encourage joint motion-frame coding and improve adaptation to hierarchical random-access structures. Experiments show that BiCRVC achieves better compression performance than state-of-the-art BVCs while providing about 30 times faster 1080p decoding than recent BVCs.

eess.IV↗

arXiv2026

Vision Language Models Cannot Plan, but Can They Formalize?

The advancement of vision language models (VLMs) has empowered embodied agents to accomplish simple multimodal planning tasks, but not long-horizon ones requiring long sequences of actions. In text-only simulations, long-horizon planning has seen significant improvement brought by repositioning the role of LLMs. Instead of directly generating action sequences, LLMs translate the planning domain and problem into a formal planning language like the Planning Domain Definition Language (PDDL), which can call a formal solver to derive the plan in a verifiable manner. In multimodal environments, research on VLM-as-formalizer remains scarce, usually involving gross simplifications such as predefined object vocabulary or overly similar few-shot examples. In this work, we present a suite of five VLM-as-formalizer pipelines that tackle one-shot, open-vocabulary, and multimodal PDDL formalization. We evaluate those on an existing benchmark while presenting another two that for the first time account for planning with authentic, multi-view, and low-quality images. We conclude that VLM-as-formalizer greatly outperforms end-to-end plan generation. We find that visual grounding of object relations remains the primary bottleneck for weaker VLMs, while stronger models have largely overcome this limitation. While generating intermediate, textual representations such as captions or scene graphs partially compensate for the performance, their inconsistent gain leaves headroom for future research directions on multimodal planning formalization.

cs.CL↗

arXiv2026

Leveraging Machine Unlearning for Cost-Efficient Preference Alignment

Despite advances in Preference Alignment (PA) for Large Language Models (LLMs), mainstream methods like reinforcement learning with human feedback face notable challenges. These approaches require high-quality datasets of positive preference examples, which are costly to obtain and computationally intensive. The LLM unlearning technique presents a promising alternative by directly removing the influence of negative examples. However, current research has primarily focused on empirical validation, lacking systematic quantitative analysis. To bridge this gap, we propose a framework linking PA with LLM unlearning. Through bi-level optimization, we first quantify how unlearning specific negative examples impacts PA performance. Our analysis reveals that these effects vary substantially across negative examples. Building on this insight, we pose a crucial question: how can we optimally select and weight negative examples for unlearning to maximize PA performance? To answer this, we propose Unlearning to Align (U2A), which leverages bi-level optimization to efficiently select and unlearn examples for optimal PA performance. We validate the proposed method through extensive experiments, with results confirming its effectiveness. Our code is available at https://github.com/muyiahhh/U2A.

cs.LG↗