arXiv subjects

Zhihao Zhang

Publications and source records attributed to Zhihao Zhang.

At least 91 records · Page 5Linked to original sources

arXiv2024

TidalDecode: Fast and Accurate LLM Decoding with Position Persistent Sparse Attention

Large language models (LLMs) have driven significant advancements across diverse NLP tasks, with long-context models gaining prominence for handling extended inputs. However, the expanding key-value (KV) cache size required by Transformer architectures intensifies the memory constraints, particularly during the decoding phase, creating a significant bottleneck. Existing sparse attention mechanisms designed to address this bottleneck have two limitations: (1) they often fail to reliably identify the most relevant tokens for attention, and (2) they overlook the spatial coherence of token selection across consecutive Transformer layers, which can lead to performance degradation and substantial overhead in token selection. This paper introduces TidalDecode, a simple yet effective algorithm and system for fast and accurate LLM decoding through position persistent sparse attention. TidalDecode leverages the spatial coherence of tokens selected by existing sparse attention methods and introduces a few token selection layers that perform full attention to identify the tokens with the highest attention scores, while all other layers perform sparse attention with the pre-selected tokens. This design enables TidalDecode to substantially reduce the overhead of token selection for sparse attention without sacrificing the quality of the generated results. Evaluation on a diverse set of LLMs and tasks shows that TidalDecode closely matches the generative performance of full attention methods while reducing the LLM decoding latency by up to 2.1x.

cs.LG↗

arXiv2024

Cold Darkogenesis: Dark Matter and Baryon Asymmetry in Light of the PTA Signal

We build upon the intriguing possibility that the recently reported nano-Hz gravitational wave signal by Pulsar Timing Array (PTA) experiments is sourced by a strong first-order phase transition from a nearly conformal dark sector. The phase transition has to be strongly supercooled to explain the signal amplitude, while the critical temperature has to be in the $\cal{O}$(GeV) range, as dictated by the peak frequency of the gravitational wave spectrum. However, the resulting strong supercooling exponentially dilutes away any pre-existing baryon asymmetry and dark matter, calling for a new paradigm of their productions. We then develop a mechanism of cold darkogenesis that generates a dark asymmetry during the phase transition from the textured dark $SU(2)_{\rm D}$ Higgs field. This dark asymmetry is transferred to the visible sector via neutron portal interactions, resulting in the observed baryon asymmetry. Furthermore, the mechanism naturally leads to the correct abundance of asymmetric dark matter, with self-interaction of the scale that is of the right order to solve the diversity problem in galactic rotation curves. Collider searches for mono-jets and dark matter direct detection experiments can dictate the viability of the model.

hep-ph↗

arXiv2024

High-order accurate structure-preserving finite volume schemes on adaptive moving meshes for shallow water equations: Well-balancedness and positivity

This paper develops high-order accurate, well-balanced (WB), and positivity-preserving (PP) finite volume schemes for shallow water equations on adaptive moving structured meshes. The mesh movement poses new challenges in maintaining the WB property, which not only depends on the balance between flux gradients and source terms but is also affected by the mesh movement. To address these complexities, the WB property in curvilinear coordinates is decomposed into flux source balance and mesh movement balance. The flux source balance is achieved by suitable decomposition of the source terms, the numerical fluxes based on hydrostatic reconstruction, and appropriate discretization of the geometric conservation laws (GCLs). Concurrently, the mesh movement balance is maintained by integrating additional schemes to update the bottom topography during mesh adjustments. The proposed schemes are rigorously proven to maintain the WB property by using the discrete GCLs and these two balances. We provide rigorous analyses of the PP property under a sufficient condition enforced by a PP limiter. Due to the involvement of mesh metrics and movement, the analyses are nontrivial, while some standard techniques, such as splitting high-order schemes into convex combinations of formally first-order PP schemes, are not directly applicable. Various numerical examples validate the high-order accuracy, high efficiency, WB, and PP properties of the proposed schemes.

math.NA↗

arXiv2024

Frequency stabilization based on H13C14N absorption in lithium niobate micro-disk laser

We demonstrate an on-chip lithium niobate micro-disk laser based on hydrogen cyanide (H13C14N) gas saturation absorption method for frequency stabilization. The laser chip consists of two main components: a micro-disk laser and a combined racetrack ring cavity. By operating on the H13C14N P12 absorption line at 1551.3 nm, the laser frequency can be precisely stabilized. The laser demonstrates remarkable stability, achieving a best stability value of 9*10^-9. Furthermore, the short-term stability, evaluated over continuous time intervals of 35 seconds, showcases exceptional performance. Additionally, the residual drift remains well below 30 MHz.

physics.optics↗

arXiv2024

HuntFUZZ: Enhancing Error Handling Testing through Clustering Based Fuzzing

Testing a program's capability to effectively handling errors is a significant challenge, given that program errors are relatively uncommon. To solve this, Software Fault Injection (SFI)-based fuzzing integrates SFI and traditional fuzzing, injecting and triggering errors for testing (error handling) code. However, we observe that current SFI-based fuzzing approaches have overlooked the correlation between paths housing error points. In fact, the execution paths of error points often share common paths. Nonetheless, Fuzzers usually generate test cases repeatedly to test error points on commonly traversed paths. This practice can compromise the efficiency of the fuzzer(s). Thus, this paper introduces HuntFUZZ, a novel SFI-based fuzzing framework that addresses the issue of redundant testing of error points with correlated paths. Specifically, HuntFUZZ clusters these correlated error points and utilizes concolic execution to compute constraints only for common paths within each cluster. By doing so, we provide the fuzzer with efficient test cases to explore related error points with minimal redundancy. We evaluate HuntFUZZ on a diverse set of 42 applications, and HuntFUZZ successfully reveals 162 known bugs, with 62 of them being related to error handling. Additionally, due to its efficient error point detection method, HuntFUZZ discovers 7 unique zero-day bugs, which are all missed by existing fuzzers. Furthermore, we compare HuntFUZZ with 4 existing fuzzing approaches, including AFL, AFL++, AFLGo, and EH-FUZZ. Our evaluation confirms that HuntFUZZ can cover a broader range of error points, and it exhibits better performance in terms of bug finding speed.

cs.CR↗

arXiv2024

An integrated electro-optically tunable multi-channel interference cavity laser

We demonstrated a continuously tunable laser system by butt coupling a reflective semiconductor optical amplifier (RSOA) chip with a thin-film lithium niobate (TFLN) based multi-channel interference (MCI) cavity chip. This hybrid integrated lasers allows for fine-tuning of the laser wavelength from 1538 nm to 1560 nm with a resolution of 0.014 nm and a side-mode suppression ratio (SMSR) exceeding 30 dB. The MCI cavity chip is fabricated using the photolithography assisted chemo-mechanical etching (PLACE) technique. The developed laser has an output power of approximately 10 μW, which can be further amplified to 70 mW using a commercial erbium-doped fiber amplifier (EDFA) without significant broadening of the laser linewidth.

physics.optics↗

arXiv2024

Generating Synthetic Computed Tomography for Radiotherapy: SynthRAD2023 Challenge Report

Radiation therapy plays a crucial role in cancer treatment, necessitating precise delivery of radiation to tumors while sparing healthy tissues over multiple days. Computed tomography (CT) is integral for treatment planning, offering electron density data crucial for accurate dose calculations. However, accurately representing patient anatomy is challenging, especially in adaptive radiotherapy, where CT is not acquired daily. Magnetic resonance imaging (MRI) provides superior soft-tissue contrast. Still, it lacks electron density information while cone beam CT (CBCT) lacks direct electron density calibration and is mainly used for patient positioning. Adopting MRI-only or CBCT-based adaptive radiotherapy eliminates the need for CT planning but presents challenges. Synthetic CT (sCT) generation techniques aim to address these challenges by using image synthesis to bridge the gap between MRI, CBCT, and CT. The SynthRAD2023 challenge was organized to compare synthetic CT generation methods using multi-center ground truth data from 1080 patients, divided into two tasks: 1) MRI-to-CT and 2) CBCT-to-CT. The evaluation included image similarity and dose-based metrics from proton and photon plans. The challenge attracted significant participation, with 617 registrations and 22/17 valid submissions for tasks 1/2. Top-performing teams achieved high structural similarity indices (>0.87/0.90) and gamma pass rates for photon (>98.1%/99.0%) and proton (>97.3%/97.0%) plans. However, no significant correlation was found between image similarity metrics and dose accuracy, emphasizing the need for dose evaluation when assessing the clinical applicability of sCT. SynthRAD2023 facilitated the investigation and benchmarking of sCT generation techniques, providing insights for developing MRI-only and CBCT-based adaptive radiotherapy.

physics.med-ph↗

arXiv2024

ATLAS: Improving Lay Summarisation with Attribute-based Control

Lay summarisation aims to produce summaries of scientific articles that are comprehensible to non-expert audiences. However, previous work assumes a one-size-fits-all approach, where the content and style of the produced summary are entirely dependent on the data used to train the model. In practice, audiences with different levels of expertise will have specific needs, impacting what content should appear in a lay summary and how it should be presented. Aiming to address this, we propose ATLAS, a novel abstractive summarisation approach that can control various properties that contribute to the overall "layness" of the generated summary using targeted control attributes. We evaluate ATLAS on a combination of biomedical lay summarisation datasets, where it outperforms state-of-the-art baselines using mainstream summarisation metrics. Additional analyses provided on the discriminatory power and emergent influence of our selected controllable attributes further attest to the effectiveness of our approach.

cs.CL↗

arXiv2024

Unveiling Linguistic Regions in Large Language Models

Large Language Models (LLMs) have demonstrated considerable cross-lingual alignment and generalization ability. Current research primarily focuses on improving LLMs' cross-lingual generalization capabilities. However, there is still a lack of research on the intrinsic mechanisms of how LLMs achieve cross-lingual alignment. From the perspective of region partitioning, this paper conducts several investigations on the linguistic competence of LLMs. We discover a core region in LLMs that corresponds to linguistic competence, accounting for approximately 1% of the total model parameters. Removing this core region by setting parameters to zero results in a significant performance decrease across 30 different languages. Furthermore, this core region exhibits significant dimensional dependence, perturbations to even a single parameter on specific dimensions leading to a loss of linguistic competence. Moreover, we discover that distinct monolingual regions exist for different languages, and disruption to these specific regions substantially reduces the LLMs' proficiency in those corresponding languages. Our research also indicates that freezing the core linguistic region during further pre-training can mitigate the issue of catastrophic forgetting (CF), a common phenomenon observed during further pre-training of LLMs. Overall, exploring the LLMs' functional regions provides insights into the foundation of their intelligence.

cs.CL↗

arXiv2024

Game-MUG: Multimodal Oriented Game Situation Understanding and Commentary Generation Dataset

The dynamic nature of esports makes the situation relatively complicated for average viewers. Esports broadcasting involves game expert casters, but the caster-dependent game commentary is not enough to fully understand the game situation. It will be richer by including diverse multimodal esports information, including audiences' talks/emotions, game audio, and game match event information. This paper introduces GAME-MUG, a new multimodal game situation understanding and audience-engaged commentary generation dataset and its strong baseline. Our dataset is collected from 2020-2022 LOL game live streams from YouTube and Twitch, and includes multimodal esports game information, including text, audio, and time-series event logs, for detecting the game situation. In addition, we also propose a new audience conversation augmented commentary dataset by covering the game situation and audience conversation understanding, and introducing a robust joint multimodal dual learning model as a baseline. We examine the model's game situation/event understanding ability and commentary generation capability to show the effectiveness of the multimodal aspects coverage and the joint integration learning approach.

cs.CL↗

arXiv2024

Adaptive Learning for Multi-view Stereo Reconstruction

Deep learning has recently demonstrated its excellent performance on the task of multi-view stereo (MVS). However, loss functions applied for deep MVS are rarely studied. In this paper, we first analyze existing loss functions' properties for deep depth based MVS approaches. Regression based loss leads to inaccurate continuous results by computing mathematical expectation, while classification based loss outputs discretized depth values. To this end, we then propose a novel loss function, named adaptive Wasserstein loss, which is able to narrow down the difference between the true and predicted probability distributions of depth. Besides, a simple but effective offset module is introduced to better achieve sub-pixel prediction accuracy. Extensive experiments on different benchmarks, including DTU, Tanks and Temples and BlendedMVS, show that the proposed method with the adaptive Wasserstein loss and the offset module achieves state-of-the-art performance.

cs.CV↗

arXiv2024

TAMM: TriAdapter Multi-Modal Learning for 3D Shape Understanding

The limited scale of current 3D shape datasets hinders the advancements in 3D shape understanding, and motivates multi-modal learning approaches which transfer learned knowledge from data-abundant 2D image and language modalities to 3D shapes. However, even though the image and language representations have been aligned by cross-modal models like CLIP, we find that the image modality fails to contribute as much as the language in existing multi-modal 3D representation learning methods. This is attributed to the domain shift in the 2D images and the distinct focus of each modality. To more effectively leverage both modalities in the pre-training, we introduce TriAdapter Multi-Modal Learning (TAMM) -- a novel two-stage learning approach based on three synergistic adapters. First, our CLIP Image Adapter mitigates the domain gap between 3D-rendered images and natural images, by adapting the visual representations of CLIP for synthetic image-text pairs. Subsequently, our Dual Adapters decouple the 3D shape representation space into two complementary sub-spaces: one focusing on visual attributes and the other for semantic understanding, which ensure a more comprehensive and effective multi-modal pre-training. Extensive experiments demonstrate that TAMM consistently enhances 3D representations for a wide range of 3D encoder architectures, pre-training datasets, and downstream tasks. Notably, we boost the zero-shot classification accuracy on Objaverse-LVIS from 46.8\% to 50.7\%, and improve the 5-way 10-shot linear probing classification accuracy on ModelNet40 from 96.1\% to 99.0\%. Project page: https://alanzhangcs.github.io/tamm-page.

cs.CV↗

arXiv2024

SpecInfer: Accelerating Generative Large Language Model Serving with Tree-based Speculative Inference and Verification

This paper introduces SpecInfer, a system that accelerates generative large language model (LLM) serving with tree-based speculative inference and verification. The key idea behind SpecInfer is leveraging small speculative models to predict the LLM's outputs; the predictions are organized as a token tree, whose nodes each represent a candidate token sequence. The correctness of all candidate token sequences represented by a token tree is verified against the LLM in parallel using a novel tree-based parallel decoding mechanism. SpecInfer uses an LLM as a token tree verifier instead of an incremental decoder, which significantly reduces the end-to-end latency and computational requirement for serving generative LLMs while provably preserving model quality. Our evaluation shows that SpecInfer outperforms existing LLM serving systems by 1.5-2.8x for distributed LLM inference and by 2.6-3.5x for offloading-based LLM inference, while preserving the same generative performance. SpecInfer is publicly available at https://github.com/flexflow/FlexFlow/

cs.CL↗

arXiv2024

Advancing Translation Preference Modeling with RLHF: A Step Towards Cost-Effective Solution

Faithfulness, expressiveness, and elegance is the constant pursuit in machine translation. However, traditional metrics like \textit{BLEU} do not strictly align with human preference of translation quality. In this paper, we explore leveraging reinforcement learning with human feedback (\textit{RLHF}) to improve translation quality. It is non-trivial to collect a large high-quality dataset of human comparisons between translations, especially for low-resource languages. To address this issue, we propose a cost-effective preference learning strategy, optimizing reward models by distinguishing between human and machine translations. In this manner, the reward model learns the deficiencies of machine translation compared to human and guides subsequent improvements in machine translation. Experimental results demonstrate that \textit{RLHF} can effectively enhance translation quality and this improvement benefits other translation directions not trained with \textit{RLHF}. Further analysis indicates that the model's language capabilities play a crucial role in preference learning. A reward model with strong language capabilities can more sensitively learn the subtle differences in translation quality and align better with real human translation preferences.

cs.CL↗

arXiv2024

Accelerating Retrieval-Augmented Language Model Serving with Speculation

Retrieval-augmented language models (RaLM) have demonstrated the potential to solve knowledge-intensive natural language processing (NLP) tasks by combining a non-parametric knowledge base with a parametric language model. Instead of fine-tuning a fully parametric model, RaLM excels at its low-cost adaptation to the latest data and better source attribution mechanisms. Among various RaLM approaches, iterative RaLM delivers a better generation quality due to a more frequent interaction between the retriever and the language model. Despite the benefits, iterative RaLM usually encounters high overheads due to the frequent retrieval step. To this end, we propose RaLMSpec, a speculation-inspired framework that provides generic speed-up over iterative RaLM while preserving the same model outputs through speculative retrieval and batched verification. By further incorporating prefetching, optimal speculation stride scheduler, and asynchronous verification, RaLMSpec can automatically exploit the acceleration potential to the fullest. For naive iterative RaLM serving, extensive evaluations over three language models on four downstream QA datasets demonstrate that RaLMSpec can achieve a speed-up ratio of 1.75-2.39x, 1.04-1.39x, and 1.31-1.77x when the retriever is an exact dense retriever, approximate dense retriever, and sparse retriever respectively compared with the baseline. For KNN-LM serving, RaLMSpec can achieve a speed-up ratio up to 7.59x and 2.45x when the retriever is an exact dense retriever and approximate dense retriever, respectively, compared with the baseline.

cs.LG↗

arXiv2024

LLaMA Beyond English: An Empirical Study on Language Capability Transfer

In recent times, substantial advancements have been witnessed in large language models (LLMs), exemplified by ChatGPT, showcasing remarkable proficiency across a range of complex tasks. However, many mainstream LLMs (e.g. LLaMA) are pretrained on English-dominant corpus, which limits their performance in other non-English languages. In this paper, we focus on how to effectively transfer the capabilities of language generation and following instructions to a non-English language. To answer this question, we conduct an extensive empirical investigation based on LLaMA, accumulating over 1440 GPU hours. We analyze the impact of key factors such as vocabulary extension, further pretraining, and instruction tuning on transfer. To accurately assess the model's level of knowledge, we employ four widely used standardized testing benchmarks: C-Eval, MMLU, AGI-Eval, and GAOKAO-Bench. Furthermore, a comprehensive evaluation of the model's response quality is conducted, considering aspects such as accuracy, fluency, informativeness, logical coherence, and harmlessness, based on LLM-Eval, a benchmarks consisting instruction tasks from 17 diverse categories. Our evaluation results demonstrate that comparable performance to state-of-the-art transfer models can be achieved with less than 1% of the pretraining data, both in terms of knowledge alignment and response quality. Furthermore, the experimental outcomes across the thirteen low-resource languages also exhibit similar trends. We anticipate that the conclusions revealed by the experiments will aid the community in developing non-English LLMs.

cs.CL↗

arXiv2024

On-chip wavelength division multiplexing by angled multimode interferometer fabricated on erbium-doped thin film lithium niobate on insulator

Photonic integrated circuits based on erbium doped thin film lithium niobate on insulator has attracted broad interests with insofar various waveguide amplifiers and microlasers demonstrated. Wideband operation facilitated by the broadband absorption and emission of erbium ions necessitates the functional integration of wavelength filter and multiplexer on the same chip. Here a low-loss wavelength division multiplexer at the resonant pumping and emission wavelengths (~1480 nm and 1530~1560 nm) of erbium ions based on angled multimode interferometer, is realized in the erbium doped thin film lithium niobate on insulator fabricated by the photolithography assisted chemomechanical etching technique. The minimum on-chip insertion losses of the fabricated device are <0.7 dB for both wavelength ranges, and a 3-dB bandwidth of >20 nm is measured at the telecom C-band. Besides, direct visualization of the multimode interference pattern by the visible upconversion fluorescence of erbium ions compares well with the simulated light propagation in the multimode interferometer. Spectral tuning of the wavelength division multiplexer by structural design is also demonstrated and discussed.

physics.optics↗

arXiv2023

Language Model as an Annotator: Unsupervised Context-aware Quality Phrase Generation

Phrase mining is a fundamental text mining task that aims to identify quality phrases from context. Nevertheless, the scarcity of extensive gold labels datasets, demanding substantial annotation efforts from experts, renders this task exceptionally challenging. Furthermore, the emerging, infrequent, and domain-specific nature of quality phrases presents further challenges in dealing with this task. In this paper, we propose LMPhrase, a novel unsupervised context-aware quality phrase mining framework built upon large pre-trained language models (LMs). Specifically, we first mine quality phrases as silver labels by employing a parameter-free probing technique called Perturbed Masking on the pre-trained language model BERT (coined as Annotator). In contrast to typical statistic-based or distantly-supervised methods, our silver labels, derived from large pre-trained language models, take into account rich contextual information contained in the LMs. As a result, they bring distinct advantages in preserving informativeness, concordance, and completeness of quality phrases. Secondly, training a discriminative span prediction model heavily relies on massive annotated data and is likely to face the risk of overfitting silver labels. Alternatively, we formalize phrase tagging task as the sequence generation problem by directly fine-tuning on the Sequence-to-Sequence pre-trained language model BART with silver labels (coined as Generator). Finally, we merge the quality phrases from both the Annotator and Generator as the final predictions, considering their complementary nature and distinct characteristics. Extensive experiments show that our LMPhrase consistently outperforms all the existing competitors across two different granularity phrase mining tasks, where each task is tested on two different domain datasets.

cs.CL↗