arXiv subjects

Guanchen Li

Publications and source records attributed to Guanchen Li.

15 recordsLinked to original sources

arXiv2026

Transport-defined photodetection through electrically selectable nonequilibrium carrier transport

Broadband optical fields can change in both intensity and spectral distribution, but a fixed-response detector maps this evolving information onto a single electrical signal. Here we demonstrate transport-defined photodetection, in which electrical bias selects how photoexcited carriers are redistributed, escape and are collected, creating complementary response functions within one shared active region. In a GaAs/AlGaAs semiconductor ratchet, light-driven ratchet transport defines the response at 0 V, whereas the spectral evolution at -2 V is consistent with field-assisted hot-carrier transport. The overlapping states span a measured spectral range of 0.4-94.5 {\mu}m at 5 K, provide state-dependent calculated detection floors and support joint infrared operation at 30 K. Direct optical beat notes at 7.432 GHz in the mid-infrared and 17.103 GHz in the terahertz demonstrate optical-to-electrical conversion. Their common-path outputs recover an imposed spatial temperature gradient and the transient field-of-view-integrated effective radiation temperature of laser-excited graphite. These results establish post-photoexcitation transport as a function-defining design variable for semiconductor photodetectors, complementing structure-defined, field-tuned and optically encoded approaches to reconfigurable photodetection.

physics.optics↗

arXiv2026

Beyond the Target: From Imitation to Collaboration in Speculative Decoding

Speculative decoding (SPD) accelerates large language model (LLM) inference by letting a smaller draft model propose multiple future tokens that are verified in parallel by a larger target model. The dominant SPD paradigm treats the target model as the sole reliable teacher, accepting a draft token only when it exactly matches the target prediction. This design implicitly assumes that the target is always the better choice at every position. In practice, this assumption does not hold. Although the draft is the weaker model overall, it is not uniformly inferior at the token level. In a meaningful fraction of cases where draft and target disagree, the draft's choice is the one that leads to the correct final answer. Inspired by this, we introduce \textbf{Collaborative Speculative Decoding (CoSpec)}, a generalization of SPD that no longer treats the target model as the sole token-level authority. CoSpec trains an arbitration policy via reinforcement learning to decide whether to accept tokens from the draft or target model, selectively accepting draft tokens at mismatches when doing so is likely to yield a correct final answer. Experimental results show that CoSpec maintains substantial speedups while surpassing target-only performance. By shifting the emphasis from imitation to collaboration, CoSpec suggests a new perspective on speculative decoding.

cs.CL↗

arXiv2026

Learnable Permutation for Structured Sparsity on Transformer Models

Structured sparsity has emerged as a popular model pruning technique, widely adopted in various architectures, including CNNs, Transformer models, and especially large language models (LLMs) in recent years. A promising direction to further improve post-pruning performance is weight permutation, which reorders model weights into patterns more amenable to pruning. However, the exponential growth of the permutation search space with the scale of Transformer architectures forces most methods to rely on greedy or heuristic algorithms, limiting the effectiveness of reordering. In this work, we propose a novel end-to-end learnable permutation framework. Our method introduces a learnable permutation cost matrix to quantify the cost of swapping any two input channels of a given weight matrix, a differentiable bipartite matching solver to obtain the optimal binary permutation matrix given a cost matrix, and a sparsity optimization loss function to directly optimize the permutation operator. We extensively validate our approach on vision and language Transformers, demonstrating that our method achieves state-of-the-art permutation results for structured sparsity.

cs.LG↗

arXiv2026

Lantern: A Minimalist Robotic Object Platform

Robotic objects are simple actuated systems that subtly blend into human environments. We design and introduce Lantern, a minimalist robotic object platform to enable building simple robotic artifacts. We conducted in-depth design and engineering iterations of Lantern's mechatronic architecture to meet specific design goals while maintaining a low build cost (~40 USD). As an extendable, open-source platform, Lantern aims to enable exploration of a range of HRI scenarios by leveraging human tendency to assign social meaning to simple forms. To evaluate Lantern's potential for HRI, we conducted a series of explorations: 1) a co-design workshop, 2) a sensory room case study, 3) distribution to external HRI labs, 4) integration into a graduate-level HRI course, and 5) public exhibitions with older adults and children. Our findings show that Lantern effectively evokes engagement, can support versatile applications ranging from emotion regulation to focused work, and serves as a viable platform for lowering barriers to HRI as a field.

cs.RO↗

arXiv2025

Training-Free Loosely Speculative Decoding: Accepting Semantically Correct Drafts Beyond Exact Match

Large language models (LLMs) achieve strong performance across diverse tasks but suffer from high inference latency due to their autoregressive generation. Speculative Decoding (SPD) mitigates this issue by verifying candidate tokens in parallel from a smaller draft model, yet its strict exact-match verification discards many semantically valid continuations. Moreover, existing training-based SPD methods often suffer from performance degradation on out-of-distribution (OOD) tasks. To this end, we propose Training-Free Loosely Speculative Decoding (FLy), a novel method that loosens the rigid verification criterion by leveraging the target model's self-corrective behavior to judge whether a draft-target mismatch remains semantically valid. FLy introduces a two-tier mechanism: an entropy-level gate that identifies whether the current token allows multiple plausible alternatives or is nearly deterministic, and a token-level deferred window that distinguishes genuine errors from differently worded yet semantically correct variants. To further reduce latency, we design a multi-level acceleration strategy that accelerates not only the target model but also the drafter itself. Owing to its training-free design, FLy composes seamlessly with arbitrary draft-target pairs and generalizes across models and domains without hyperparameter re-tuning. Experiments show that FLy preserves more than 99% of the target model's accuracy while achieving an average 2.81x speedup on Llama-3.1-70B-Instruct and 5.07x speedup on the 405B variant. Notably, on out-of-domain datasets, our method remains highly effective and outperforms the training-based method EAGLE-3 by 1.62x.

cs.CL↗

arXiv2025

SparK: Query-Aware Unstructured Sparsity with Recoverable KV Cache Channel Pruning

Long-context inference in large language models (LLMs) is increasingly constrained by the KV cache bottleneck: memory usage grows linearly with sequence length, while attention computation scales quadratically. Existing approaches address this issue by compressing the KV cache along the temporal axis through strategies such as token eviction or merging to reduce memory and computational overhead. However, these methods often neglect fine-grained importance variations across feature dimensions (i.e., the channel axis), thereby limiting their ability to effectively balance efficiency and model accuracy. In reality, we observe that channel saliency varies dramatically across both queries and positions: certain feature channels carry near-zero information for a given query, while others spike in relevance. To address this oversight, we propose SPARK, a training-free plug-and-play method that applies unstructured sparsity by pruning KV at the channel level, while dynamically restoring the pruned entries during attention score computation. Notably, our approach is orthogonal to existing KV compression and quantization techniques, making it compatible for integration with them to achieve further acceleration. By reducing channel-level redundancy, SPARK enables processing of longer sequences within the same memory budget. For sequences of equal length, SPARK not only preserves or improves model accuracy but also reduces KV cache storage by over 30% compared to eviction-based methods. Furthermore, even with an aggressive pruning ratio of 80%, SPARK maintains performance with less degradation than 5% compared to the baseline eviction method, demonstrating its robustness and effectiveness. Our code will be available at https://github.com/Xnhyacinth/SparK.

cs.CL↗

arXiv2025

T\'yr-the-Pruner: Structural Pruning LLMs via Global Sparsity Distribution Optimization

Structural pruning enhances hardware-agnostic inference efficiency for large language models (LLMs) yet often fails to maintain comparable performance. Local pruning performs efficient layer-by-layer compression but ignores global topology. Although global pruning aims to identify an optimal sparse model, intuitive methods typically adopt a two-stage paradigm that first evaluates substructure saliency and then applies global pruning, which ignores inter-structure dependencies and fails to achieve end-to-end optimization. To address these limitations, we propose T\'yr-the-Pruner, an efficient end-to-end search-based global structural pruning framework. This framework constructs a supernet by repeatedly applying local pruning across a range of sparsity ratios to each layer in an LLM, with the core goal of determining the optimal sparsity distribution under a target overall sparsity ratio. Concretely, we introduce an effective local pruning and an expectation error accumulation approach to improve supernet construction. Furthermore, we employ an iterative prune-and-search strategy with coarse-to-fine sparsity granularity to ensure efficient search convergence. Experimental results show that T\'yr-the-Pruner achieves state-of-the-art structural pruning, retaining 97% of the dense model's performance while removing a challenging 50% of Llama-3.1-70B's parameters. Code will be available at https://github.com/AMD-AGI/Tyr-the-Pruner.

cs.LG↗

arXiv2024

Enhancing One-shot Pruned Pre-trained Language Models through Sparse-Dense-Sparse Mechanism

Pre-trained language models (PLMs) are engineered to be robust in contextual understanding and exhibit outstanding performance in various natural language processing tasks. However, their considerable size incurs significant computational and storage costs. Modern pruning strategies employ one-shot techniques to compress PLMs without the need for retraining on task-specific or otherwise general data; however, these approaches often lead to an indispensable reduction in performance. In this paper, we propose SDS, a Sparse-Dense-Sparse pruning framework to enhance the performance of the pruned PLMs from a weight distribution optimization perspective. We outline the pruning process in three steps. Initially, we prune less critical connections in the model using conventional one-shot pruning methods. Next, we reconstruct a dense model featuring a pruning-friendly weight distribution by reactivating pruned connections with sparse regularization. Finally, we perform a second pruning round, yielding a superior pruned model compared to the initial pruning. Experimental results demonstrate that SDS outperforms the state-of-the-art pruning techniques SparseGPT and Wanda under an identical sparsity configuration. For instance, SDS reduces perplexity by 9.13 on Raw-Wikitext2 and improves accuracy by an average of 2.05% across multiple zero-shot benchmarks for OPT-125M with 2:4 sparsity.

cs.CL↗

arXiv2024

Amphista: Bi-directional Multi-head Decoding for Accelerating LLM Inference

Large Language Models (LLMs) inherently use autoregressive decoding, which lacks parallelism in inference and results in significantly slow inference speed. While methods such as Medusa constructs parallelized heads, they lack adequate information interaction across different prediction positions. To overcome this limitation, we introduce Amphista, an enhanced speculative decoding framework that builds upon Medusa. Specifically, Amphista models an Auto-embedding Block capable of parallel inference, incorporating bi-directional attention to enable interaction between different drafting heads. Additionally, Amphista integrates Staged Adaptation Layers, which ensure a seamless transition of semantic information from the target model's autoregressive inference to the drafting heads' non-autoregressive inference, effectively achieving paradigm shift and feature fusion. Experimental results on Vicuna models using MT-Bench and Spec-Bench demonstrate that Amphista achieves substantial acceleration while maintaining generation quality. On MT-Bench, Amphista delivers up to 2.75$\times$ speedup over vanilla autoregressive decoding and 1.40$\times$ over Medusa on Vicuna 33B in wall-clock time.

cs.AI↗

arXiv2022

On the Number of Maintenance Cycles in Systems with Critical and Non-Critical Components

We present a novel mathematical framework for computing the number of maintenance cycles in a system with critical and non-critical components, where "critical" (CR) means that the component's failure is fatal for the system's operation and renders any more repairs inapplicable, whereas "noncritical" (NC) means that the component can undergo corrective maintenance (replacement or minimal repair) whenever it fails, provided that the CR component is still in operation. Whenever the NC component fails, the CR component can optionally be preventively replaced. We extend traditional renewal theory (whether classical or generalized) for various maintenance scenarios for a system composed of one CR and one NC component in order to compute the average number of renewals of NC under the restriction ("bound") necessitated by CR. We also develop approximations in closed form for the proposed "bounded" renewal functions. We validate our formulas by simulations on a variety of component lifetime distributions, including actual lifetime distributions of wind turbine components.

cs.DM↗

arXiv2016

First principle thermodynamic study of oxygen vacancy at metal/oxide interface

The oxygen vacancy is a crucial intrinsic defect in metal-ultrathin oxide semiconductor heterostructures, and its formation at an interface is of great importance in determining the device performance and degradation. This paper presents an ab initio thermodynamic study of oxygen vacancies at metal/oxide interfaces. Electronic energies and entropies are calculated for defective interface systems, as a function of interface-vacancy distance. The study indicates that oxygen vacancies near the interface modify its bonding structure, and significantly change the thermodynamic properties of the system (i.e., electronic energy and entropy) compared to bulk-like oxygen vacancies. We illustrate that different factors, including the vacancy location dependence on the energy and entropy, the temperature dependence on the entropy, and the temperature and partial pressure dependence on the oxygen chemical potential, are all important in determining the Gibbs free energy of formation of oxygen vacancy.

cond-mat.mtrl-sci↗

arXiv2016

Hydrogen release at metal-oxide interfaces: A first principle study of hydrogenated Al/SiO$_2$ interfaces

The Anode Hydrogen Release (AHR) mechanism at interfaces is responsible for the generation of defects, that traps charge carriers and can induce dielectric breakdown in Metal-Oxide-Semiconductor Field Effect Transistors. The AHR has been extensively studied at Si/SiO$_2$ interfaces but its characteristics at metal-silica interfaces remain unclear. In this study, we performed Density Functional Theory (DFT) calculations to study the hydrogen release mechanism at the typical Al/SiO$_2$ metal-oxide interface. We found that interstitial hydrogen atoms can break interfacial Al-Si bonds, passivating a Si $sp^3$ orbital. Interstitial hydrogen atoms can also break interfacial Al-O bonds, or be adsorbed at the interface on aluminum, forming stable Al-H-Al bridges. We showed that hydrogenated O-H, Si-H and Al-H bonds at the Al/SiO$_2$ interfaces are polarized. The resulting bond dipole weakens the O-H and Si-H bonds, but strengthens the Al-H bond under the application of a positive bias at the metal gate. Our calculations indicate that Al-H bonds and O-H bonds are more important than Si-H bonds for the hydrogen release process.

cond-mat.mtrl-sci↗

arXiv2016

Atomic bonding and electrical potential at metal/oxide interfaces, a first principle study

A number of electronic devices involve metal/oxide interfaces in their structure where the oxide layer plays the role of electrical insulator. As the downscaling of devices continues, the oxide thickness can spread over only a few atomic layers, making the role of interfaces prominent on its insulating properties. The prototypical Al/SiO2 metal/oxide interface is investigated using first principle calculations, and the effect of the interfacial atomic bonding is evidenced. It is shown that the interface bonding configuration critically dictates the mechanical and electronic properties of the interface. Oxygen atoms are found to better delimit the oxide boundaries than cations. Interfacial cation-metal bonds allow the metal potential to leak inside the oxide layer, without atomic diffusion, leading to a virtual oxide thinning.

cond-mat.mtrl-sci↗

arXiv2016

Steepest-entropy-ascent quantum thermodynamic modeling of heat and mass diffusion in a far-from-equilibrium system based on a single particle ensemble

This paper presents a nonequilibrium thermodynamic model for the relaxation of a local, isolated system in nonequilibrium using the principle of steepest entropy ascent (SEA), which can be expressed as a variational principle in thermodynamic state space. The model is able to arrive at the Onsager relations for such a system. Since no assumption of local equilibrium is made, the conjugate fluxes and forces, which result, are intrinsic to the subspaces of the system's state space and are defined using the concepts of hypoequilibrium state and nonequilibrium intensive properties, which describe the non-mutual equilibrium status between subspaces of the thermodynamic state space. The Onsager relations are shown to be a thermodynamic kinematic feature of the system independent of the specific details of the micro-mechanical dynamics. Two kinds of relaxation processes are studied with different constraints (i.e., conservation laws) corresponding to heat and mass diffusion. Linear behavior in the near-equilibrium region as well as nonlinear behavior in the far-from-equilibrium region are discussed. Thermodynamic relations in the equilibrium and near-equilibrium realm, including the Gibbs relation, the Clausius inequality, and the Onsager relations, are generalized to the far-from-equilibrium realm. The variational principle in the space spanned by the intrinsic conjugate fluxes and forces is expressed via the quadratic dissipation potential. As an application, the model is applied to the heat and mass diffusion of a system represented by a single particle ensemble, which can also be applied to a simple system of many particles. Phenomenological transport coefficients are also derived in near-equilibrium realm.

cond-mat.stat-mech↗

arXiv2016

Steepest-entropy-ascent quantum thermodynamic modeling of the far-from-equilibrium interactions between nonequilibrium systems of indistinguishable particle ensembles

This paper presents a nonequilibrium, first-principles, thermodynamic-ensemble based model for the relaxation process of interacting non-equilibrium systems. This model is formulated using steepest-entropy-ascent quantum thermodynamics (SEAQT) and its equation of motion for a grand canonical ensemble and is applied to a many particle system of classical or indistinguishable particles. Two kinds of interactions are discussed, including pure heat diffusion and heat and mass diffusion together. Since no local equilibrium assumption is made, the conjugate fluxes and forces are intrinsic to the subspaces of the state space of one system and/or of the state space of the two interacting systems. They are derived via the concepts of hypoequilibrium state and nonequilibrium intensive properties, which describe the nonmutual equilibrium status between subspaces of the thermodynamic state space of a single system and/or of the state space of the two interacting systems. The Onsager relations are shown to be thermodynamic kinematic features of the system and are found without knowledge of the detailed mechanics of the dynamic process. A fundamental thermodynamic explanation for the measurement of each intensive property of a system in a nonequilibrium state is given. The fundamental thermodynamic definition of reservoir is also discussed. Finally, the equation of motion for a system undergoing multiple interactions is provided, which permits the modeling of a network of local systems in nonequilibrium at any spatial and temporal scale.

cond-mat.stat-mech↗