Search

This paper proposes a novel approach for designing Single-Parameterized Kolmogorov-Arnold Networks (SKAN) by utilizing a Single-Parameterized Function (SFunc) constructed from trigonometric functions. Three new SKAN variants are developed: LSin-SKAN, LCos-SKAN, and LArctan-SKAN. Experimental validation on the MNIST dataset demonstrates that LArctan-SKAN excels in both accuracy and computational efficiency. Specifically, LArctan-SKAN significantly improves test set accuracy over existing models, outperforming all pure KAN variants compared, including FourierKAN, LSS-SKAN, and Spl-KAN. It also surpasses mixed MLP-based models such as MLP+rKAN and MLP+fKAN in accuracy. Furthermore, LArctan-SKAN exhibits remarkable computational efficiency, with a training speed increase of 535.01% and 49.55% compared to MLP+rKAN and MLP+fKAN, respectively. These results confirm the effectiveness and potential of SKANs constructed with trigonometric functions. The experiment code is available at https://github.com/chikkkit/LArctan-SKAN ., Comment: 7 pages, 3 figures, experiment code is available at https://github.com/chikkkit/LArctan-SKAN

The last year has witnessed the rapid progress of large language models (LLMs) across diverse domains. Among them, CodeLLMs have garnered particular attention because they can not only assist in completing various programming tasks but also represent the decision-making and logical reasoning capabilities of LLMs. However, current CodeLLMs mainly focus on pre-training and supervised fine-tuning scenarios, leaving the alignment stage, which is important for post-training LLMs, under-explored. This work first identifies that the commonly used PPO algorithm may be suboptimal for the alignment of CodeLLM because the involved reward rules are routinely coarse-grained and potentially flawed. We then advocate addressing this using the DPO algorithm. Based on only preference data pairs, DPO can render the model rank data automatically, giving rise to a fine-grained rewarding pattern more robust than human intervention. We also contribute a pipeline for collecting preference pairs for DPO on CodeLLMs. Studies show that our method significantly improves the performance of existing CodeLLMs on benchmarks such as MBPP and HumanEval.

Multimodal Large Language Models (MLLMs) utilize multimodal contexts consisting of text, images, or videos to solve various multimodal tasks. However, we find that changing the order of multimodal input can cause the model's performance to fluctuate between advanced performance and random guessing. This phenomenon exists in both single-modality (text-only or image-only) and mixed-modality (image-text-pair) contexts. Furthermore, we demonstrate that popular MLLMs pay special attention to certain multimodal context positions, particularly the beginning and end. Leveraging this special attention, we place key video frames and important image/text content in special positions within the context and submit them to the MLLM for inference. This method results in average performance gains of 14.7% for video-caption matching and 17.8% for visual question answering tasks. Additionally, we propose a new metric, Position-Invariant Accuracy (PIA), to address order bias in MLLM evaluation. Our research findings contribute to a better understanding of Multi-Modal In-Context Learning (MMICL) and provide practical strategies for enhancing MLLM performance without increasing computational costs.

Topological defects are discontinuities of a system protected by global properties, with wide applications in mathematics and physics. While previous experimental studies mostly focused on their classical properties, it has been predicted that topological defects can exhibit quantum superposition. Despite the fundamental interest and potential applications in understanding symmetry-breaking dynamics of quantum phase transitions, its experimental realization still remains a challenge. Here, we report the observation of quantum superposition of topological defects in a trapped-ion quantum simulator. By engineering long-range spin-spin interactions, we observe a spin kink splitting into a superposition of kinks at different positions, creating a ``Schrodinger kink'' that manifests non-locality and quantum interference. Furthermore, by preparing superposition states of neighboring kinks with different phases, we observe the propagation of the wave packet in different directions, thus unambiguously verifying the quantum coherence in the superposition states. Our work provides useful tools for non-equilibrium dynamics in quantum Kibble-Zurek physics., Comment: 8 pages, 6 figures, already published in Science Advances

Text embeddings are vital for tasks such as text retrieval and semantic textual similarity (STS). Recently, the advent of pretrained language models, along with unified benchmarks like the Massive Text Embedding Benchmark (MTEB), has facilitated the development of versatile general-purpose text embedding models. Advanced embedding models are typically developed using large-scale multi-task data and joint training across multiple tasks. However, our experimental analysis reveals two significant drawbacks of joint training: 1) Task Conflict: Gradients from different tasks interfere with each other, leading to negative transfer. 2) Data Imbalance: Disproportionate data distribution introduces biases that negatively impact performance across tasks. To overcome these challenges, we explore model merging-a technique that combines independently trained models to mitigate gradient conflicts and balance data distribution. We introduce a novel method, Self Positioning, which efficiently searches for optimal model combinations within the interpolation space of task vectors using stochastic gradient descent. Our experiments demonstrate that Self Positioning significantly enhances multi-task performance on the MTEB dataset, achieving an absolute improvement of 0.7 points. It outperforms traditional resampling methods while reducing computational costs. This work offers a robust approach to building generalized text embedding models with superior performance across diverse embedding-related tasks., Comment: working in progress

The recently proposed Kolmogorov-Arnold Networks (KAN) networks have attracted increasing attention due to their advantage of high visualizability compared to MLP. In this paper, based on a series of small-scale experiments, we proposed the Efficient KAN Expansion Principle (EKE Principle): allocating parameters to expand network scale, rather than employing more complex basis functions, leads to more efficient performance improvements in KANs. Based on this principle, we proposed a superior KAN termed SKAN, where the basis function utilizes only a single learnable parameter. We then evaluated various single-parameterized functions for constructing SKANs, with LShifted Softplus-based SKANs (LSS-SKANs) demonstrating superior accuracy. Subsequently, extensive experiments were performed, comparing LSS-SKAN with other KAN variants on the MNIST dataset. In the final accuracy tests, LSS-SKAN exhibited superior performance on the MNIST dataset compared to all tested pure KAN variants. Regarding execution speed, LSS-SKAN outperformed all compared popular KAN variants. Our experimental codes are available at https://github.com/chikkkit/LSS-SKAN and SKAN's Python library (for quick construction of SKAN in python) codes are available at https://github.com/chikkkit/SKAN ., Comment: 25 pages, 14 figures, experiment codes are available at https://github.com/chikkkit/LSS-SKAN , and SKAN's Python library code are available at https://github.com/chikkkit/SKAN

Recent advancements in building domain-specific large language models (LLMs) have shown remarkable success, especially in tasks requiring reasoning abilities like logical inference over complex relationships and multi-step problem solving. However, creating a powerful all-in-one LLM remains challenging due to the need for proprietary data and vast computational resources. As a resource-friendly alternative, we explore the potential of merging multiple expert models into a single LLM. Existing studies on model merging mainly focus on generalist LLMs instead of domain experts, or the LLMs under the same architecture and size. In this work, we propose an unconstrained model merging framework that accommodates both homogeneous and heterogeneous model architectures with a focus on reasoning tasks. A fine-grained layer-wise weight merging strategy is designed for homogeneous models merging, while heterogeneous model merging is built upon the probabilistic distribution knowledge derived from instruction-response fine-tuning data. Across 7 benchmarks and 9 reasoning-optimized LLMs, we reveal key findings that combinatorial reasoning emerges from merging which surpasses simple additive effects. We propose that unconstrained model merging could serve as a foundation for decentralized LLMs, marking a notable progression from the existing centralized LLM framework. This evolution could enhance wider participation and stimulate additional advancement in the field of artificial intelligence, effectively addressing the constraints posed by centralized models., Comment: Under review, correct typos

KV cache has become a de facto technique for the inference of large language models (LLMs), where tensors of shape (layer number, head number, sequence length, feature dimension) are introduced to cache historical information for self-attention. As the size of the model and data grows, the KV cache can quickly become a bottleneck within the system in both storage and memory transfer. To address this, prior studies usually focus on the first three axes of the cache tensors for compression. This paper supplements them, focusing on the feature dimension axis, by utilizing low-rank projection matrices to transform the cache features into spaces with reduced dimensions. We begin by investigating the canonical orthogonal projection method for data compression through principal component analysis (PCA). We observe the issue with PCA projection where significant performance degradation is observed at low compression rates. To bridge the gap, we propose to directly tune the orthogonal projection matrices with a distillation objective using an elaborate Matryoshka training strategy. After training, we adaptively search for the optimal compression rates for various layers and heads given varying compression budgets. Compared to previous works, our method can easily embrace pre-trained LLMs and hold a smooth tradeoff between performance and compression rate. We empirically witness the high data efficiency of our training procedure and find that our method can sustain over 90% performance with an average KV cache compression rate of 60% (and up to 75% in certain extreme scenarios) for popular LLMs like LLaMA2-7B-base and Mistral-7B-v0.3-base.

This paper explores testing unit roots based on least absolute deviations (LAD) regression under unconditional heteroskedasticity. We first derive the asymptotic properties of the LAD estimator for a first-order autoregressive process with the coefficient (local to) unity under unconditional heteroskedasticity and weak dependence, revealing that the limiting distribution of the LAD estimator (consequently the derived test statistics) is closely associated with unknown time-varying variances. To conduct feasible LAD-based unit root tests under heteroskedasticity and serial dependence, we develop an adaptive block bootstrap procedure, which accommodates time-varying volatility and serial dependence, both of unknown forms, to compute critical values for LAD-based tests. The asymptotic validity is established. We then extend the testing procedure to allow for deterministic components. Simulation results indicate that, in the presence of unconditional heteroskedasticity and serial dependence, the classic LAD-based tests demonstrate severe size distortion, whereas the proposed LAD-based bootstrap tests exhibit good size-control capability. Additionally, the newly developed tests show superior testing power in heavy-tailed distributed cases compared to considered benchmarks. Finally, empirical analysis of real effective exchange rates of 16 EU countries is conducted to illustrate the applicability of the newly proposed tests.

With the rapid progress of diffusion-based content generation, significant efforts are being made to unlearn harmful or copyrighted concepts from pretrained diffusion models (DMs) to prevent potential model misuse. However, it is observed that even when DMs are properly unlearned before release, malicious finetuning can compromise this process, causing DMs to relearn the unlearned concepts. This occurs partly because certain benign concepts (e.g., "skin") retained in DMs are related to the unlearned ones (e.g., "nudity"), facilitating their relearning via finetuning. To address this, we propose meta-unlearning on DMs. Intuitively, a meta-unlearned DM should behave like an unlearned DM when used as is; moreover, if the meta-unlearned DM undergoes malicious finetuning on unlearned concepts, the related benign concepts retained within it will be triggered to self-destruct, hindering the relearning of unlearned concepts. Our meta-unlearning framework is compatible with most existing unlearning methods, requiring only the addition of an easy-to-implement meta objective. We validate our approach through empirical experiments on meta-unlearning concepts from Stable Diffusion models (SD-v1-4 and SDXL), supported by extensive ablation studies. Our code is available at https://github.com/sail-sg/Meta-Unlearning.

The KV-Cache technique has become the standard for the inference of large language models (LLMs). It caches states of self-attention to avoid recomputation. Yet, it is widely criticized that KV-Cache can become a bottleneck of the LLM inference system, especially when confronted with ultra-large models and long-context queries. A natural remedy is to discard the KV-Cache for less important tokens, with StreamingLLM as an example, but the used static eviction strategies cannot flexibly adapt to varying contexts. Remedies like H2O leverage accumulative attention scores to perform dynamic eviction but suffer from the attention bias issue in capturing contextual information. This paper bridges this gap by devising a parameterized KV-Cache eviction mechanism, dubbed as Attention-Gate, which accepts the whole context as input and yields eviction flags for each token to realize in-context eviction. The subsequent self-attention module proceeds according to the flags and only the KV states for the remaining tokens need to be cached. The Attention-Gates can vary among different heads and layers and be trivially plugged into pre-trained LLMs, tuned by cost-effective continual pre-training or supervised fine-tuning objectives to acquire what to discard. The computational and memory overhead introduced by Attention-Gates is minimal. Our method is validated across multiple tasks, demonstrating both efficiency and adaptability. After a highly efficient continual pre-training, it achieves higher average accuracy and evicts more tokens compared to traditional training-free methods. In supervised fine-tuning, it not only evicts many tokens but also outperforms LoRA-finetuned LLMs on some datasets, such as RTE, where it improves accuracy by 13.9% while evicting 62.8% of tokens, showing that effective eviction of redundant tokens can even enhance performance.

Enabling mobile robots to perform long-term tasks in dynamic real-world environments is a formidable challenge, especially when the environment changes frequently due to human-robot interactions or the robot's own actions. Traditional methods typically assume static scenes, which limits their applicability in the continuously changing real world. To overcome these limitations, we present DovSG, a novel mobile manipulation framework that leverages dynamic open-vocabulary 3D scene graphs and a language-guided task planning module for long-term task execution. DovSG takes RGB-D sequences as input and utilizes vision-language models (VLMs) for object detection to obtain high-level object semantic features. Based on the segmented objects, a structured 3D scene graph is generated for low-level spatial relationships. Furthermore, an efficient mechanism for locally updating the scene graph, allows the robot to adjust parts of the graph dynamically during interactions without the need for full scene reconstruction. This mechanism is particularly valuable in dynamic environments, enabling the robot to continually adapt to scene changes and effectively support the execution of long-term tasks. We validated our system in real-world environments with varying degrees of manual modifications, demonstrating its effectiveness and superior performance in long-term tasks. Our project page is available at: https://bjhyzj.github.io/dovsg-web., Comment: 8 pages, 5 figures

This paper presents EasyRAG, a simple, lightweight, and efficient retrieval-augmented generation framework for automated network operations. Our framework has three advantages. The first is accurate question answering. We designed a straightforward RAG scheme based on (1) a specific data processing workflow (2) dual-route sparse retrieval for coarse ranking (3) LLM Reranker for reranking (4) LLM answer generation and optimization. This approach achieved first place in the GLM4 track in the preliminary round and second place in the GLM4 track in the semifinals. The second is simple deployment. Our method primarily consists of BM25 retrieval and BGE-reranker reranking, requiring no fine-tuning of any models, occupying minimal VRAM, easy to deploy, and highly scalable; we provide a flexible code library with various search and generation strategies, facilitating custom process implementation. The last one is efficient inference. We designed an efficient inference acceleration scheme for the entire coarse ranking, reranking, and generation process that significantly reduces the inference latency of RAG while maintaining a good level of accuracy; each acceleration scheme can be plug-and-play into any component of the RAG process, consistently enhancing the efficiency of the RAG system. Our code and data are released at \url{https://github.com/BUAADreamer/EasyRAG}., Comment: 10 pages, 2 figures

The efficacy of video generation models heavily depends on the quality of their training datasets. Most previous video generation models are trained on short video clips, while recently there has been increasing interest in training long video generation models directly on longer videos. However, the lack of such high-quality long videos impedes the advancement of long video generation. To promote research in long video generation, we desire a new dataset with four key features essential for training long video generation models: (1) long videos covering at least 10 seconds, (2) long-take videos without cuts, (3) large motion and diverse contents, and (4) temporally dense captions. To achieve this, we introduce a new pipeline for selecting high-quality long-take videos and generating temporally dense captions. Specifically, we define a set of metrics to quantitatively assess video quality including scene cuts, dynamic degrees, and semantic-level quality, enabling us to filter high-quality long-take videos from a large amount of source videos. Subsequently, we develop a hierarchical video captioning pipeline to annotate long videos with temporally-dense captions. With this pipeline, we curate the first long-take video dataset, LVD-2M, comprising 2 million long-take videos, each covering more than 10 seconds and annotated with temporally dense captions. We further validate the effectiveness of LVD-2M by fine-tuning video generation models to generate long videos with dynamic motions. We believe our work will significantly contribute to future research in long video generation., Comment: NeurIPS 2024 Dataset and Benchmark Track. Project page: https://silentview.github.io/LVD-2M/ . Code: https://github.com/SilentView/LVD-2M

Building on the advancements of Large Language Models (LLMs) and Vision Language Models (VLMs), recent research has introduced Vision-Language-Action (VLA) models as an integrated solution for robotic manipulation tasks. These models take camera images and natural language task instructions as input and directly generate control actions for robots to perform specified tasks, greatly improving both decision-making capabilities and interaction with human users. However, the data-driven nature of VLA models, combined with their lack of interpretability, makes the assurance of their effectiveness and robustness a challenging task. This highlights the need for a reliable testing and evaluation platform. For this purpose, in this work, we propose LADEV, a comprehensive and efficient platform specifically designed for evaluating VLA models. We first present a language-driven approach that automatically generates simulation environments from natural language inputs, mitigating the need for manual adjustments and significantly improving testing efficiency. Then, to further assess the influence of language input on the VLA models, we implement a paraphrase mechanism that produces diverse natural language task instructions for testing. Finally, to expedite the evaluation process, we introduce a batch-style method for conducting large-scale testing of VLA models. Using LADEV, we conducted experiments on several state-of-the-art VLA models, demonstrating its effectiveness as a tool for evaluating these models. Our results showed that LADEV not only enhances testing efficiency but also establishes a solid baseline for evaluating VLA models, paving the way for the development of more intelligent and advanced robotic systems., Comment: 8 pages, 4 figures

It is desirable but challenging to generate content-rich long videos in the scale of minutes. Autoregressive large language models (LLMs) have achieved great success in generating coherent and long sequences of tokens in the domain of natural language processing, while the exploration of autoregressive LLMs for video generation is limited to generating short videos of several seconds. In this work, we conduct a deep analysis of the challenges that prevent autoregressive LLM-based video generators from generating long videos. Based on the observations and analysis, we propose Loong, a new autoregressive LLM-based video generator that can generate minute-long videos. Specifically, we model the text tokens and video tokens as a unified sequence for autoregressive LLMs and train the model from scratch. We propose progressive short-to-long training with a loss re-weighting scheme to mitigate the loss imbalance problem for long video training. We further investigate inference strategies, including video token re-encoding and sampling strategies, to diminish error accumulation during inference. Our proposed Loong can be trained on 10-second videos and be extended to generate minute-level long videos conditioned on text prompts, as demonstrated by the results. More samples are available at: https://epiphqny.github.io/Loong-video., Comment: Project page: https://epiphqny.github.io/Loong-video/

Scanning Probe Microscopy (SPM) techniques have shown great potential in fabricating nanoscale structures endowed with exotic quantum properties achieved through various manipulations of atoms and molecules. However, the selection of proper manipulation parameters requires extensive domain knowledge, which is not necessarily transferable to new systems. Therefore, efficient and autonomous SPM techniques are needed to reduce the reliance on user supervision and learn optimal strategies for new systems, in particular for the challenge of controlling chemical reactions. In this paper, we developed a software infrastructure named AutoOSS (Autonomous On-Surface Synthesis) to automate bromine removal from Zn(II)-5,15- bis(4-bromo-2,6-dimethylphenyl)porphyrin (ZnBr2Me4DPP) on Au(111), using neural network models to interpret STM outputs and deep reinforcement learning models to optimize manipulation parameters. This is further supported by Bayesian Optimization Structure Search (BOSS) and Density Functional Theory (DFT) computations to explore 3D structures and reaction mechanisms based on STM images.

Artificial intelligence is gradually demonstrating its immense potential, and increasing attention is being given to how AI can be harnessed to advance scientific research. In this vision paper, we present our perspectives on how AI can better assist scientific inquiry and explore corresponding technical approach. We have proposed and open-sourced a large model of our KALE-LM model series, Llama3-KALE-LM-Chem-8B, which has achieved outstanding performance in tasks related to the field of chemistry. We hope that our work serves as a strong starting point, helping to realize more intelligent AI and promoting the advancement of human science and technology, as well as societal development.

In this study, we delve into the efficacy of transformers within pre-trained language models (PLMs) when repurposed as encoders for Automatic Speech Recognition (ASR). Our underlying hypothesis posits that, despite being initially trained on text-based corpora, these transformers possess a remarkable capacity to extract effective features from the input sequence. This inherent capability, we argue, is transferrable to speech data, thereby augmenting the acoustic modeling ability of ASR. Through rigorous empirical analysis, our findings reveal a notable improvement in Character Error Rate (CER) and Word Error Rate (WER) across diverse ASR tasks when transformers from pre-trained LMs are incorporated. Particularly, they serve as an advantageous starting point for initializing ASR encoders. Furthermore, we uncover that these transformers, when integrated into a well-established ASR encoder, can significantly boost performance, especially in scenarios where profound semantic comprehension is pivotal. This underscores the potential of leveraging the semantic prowess embedded within pre-trained transformers to advance ASR systems' capabilities., Comment: 8pages

Sequential recommendation represents a pivotal branch of recommendation systems, centered around dynamically analyzing the sequential dependencies between user preferences and their interactive behaviors. Despite the Transformer architecture-based models achieving commendable performance within this domain, their quadratic computational complexity relative to the sequence dimension impedes efficient modeling. In response, the innovative Mamba architecture, characterized by linear computational complexity, has emerged. Mamba4Rec further pioneers the application of Mamba in sequential recommendation. Nonetheless, Mamba 1's hardware-aware algorithm struggles to efficiently leverage modern matrix computational units, which lead to the proposal of the improved State Space Duality (SSD), also known as Mamba 2. While the SSD4Rec successfully adapts the SSD architecture for sequential recommendation, showing promising results in high-dimensional contexts, it suffers significant performance drops in low-dimensional scenarios crucial for pure ID sequential recommendation tasks. Addressing this challenge, we propose a novel sequential recommendation backbone model, TiM4Rec, which ameliorates the low-dimensional performance loss of the SSD architecture while preserving its computational efficiency. Drawing inspiration from TiSASRec, we develop a time-aware enhancement method tailored for the linear computation demands of the SSD architecture, thereby enhancing its adaptability and achieving state-of-the-art (SOTA) performance in both low and high-dimensional modeling. The code for our model is publicly accessible at https://github.com/AlwaysFHao/TiM4Rec.

Multi-modal foundation models and generative AI have demonstrated promising capabilities in applications across various domains. Recently, Vision-language-action (VLA) models have attracted much attention regarding their potential to advance robotic manipulation. Despite the end-to-end perception-control loop offered by the VLA models, there is a lack of comprehensive understanding of the capabilities of such models and an automated testing platform to reveal their robustness and reliability across different robotic manipulation scenarios. To address these challenges, in this work, we present VLATest, a testing framework that automatically generates diverse robotic manipulation scenes to assess the performance of VLA models from various perspectives. Large-scale experiments are considered, including eight VLA models, four types of manipulation tasks, and over 18,604 testing scenes. The experimental results show that existing VAL models still lack imperative robustness for practical applications. Specifically, the performance of VLA models can be significantly affected by several factors from the operation environments, such as camera poses, lighting conditions, and unseen objects. Our framework and the insights derived from the study are expected to pave the way for more advanced and reliable VLA-enabled robotic manipulation systems in practice., Comment: 14 pages, 7 figures

Weight normalization (WeightNorm) is widely used in practice for the training of deep neural networks and modern deep learning libraries have built-in implementations of it. In this paper, we provide the first theoretical characterizations of both optimization and generalization of deep WeightNorm models with smooth activation functions. For optimization, from the form of the Hessian of the loss, we note that a small Hessian of the predictor leads to a tractable analysis. Thus, we bound the spectral norm of the Hessian of WeightNorm networks and show its dependence on the network width and weight normalization terms--the latter being unique to networks without WeightNorm. Then, we use this bound to establish training convergence guarantees under suitable assumptions for gradient decent. For generalization, we use WeightNorm to get a uniform convergence based generalization bound, which is independent from the width and depends sublinearly on the depth. Finally, we present experimental results which illustrate how the normalization terms and other quantities of theoretical interest relate to the training of WeightNorm networks.

The two-dimensional molecular conductor $\kappa$-(BETS)$_2$Mn[N(CN)$_2$]$_3$ undergoes a sharp metal-to-insulator phase transition at $T_{\rm MI}\approx$ 21 K, which has been under scrutiny for many years. We have performed comprehensive infrared investigations along the three crystallographic directions as a function of temperature down to 10 K, complemented by electron spin resonance and dc-transport studies. The in-plane anisotropy of the optical conductivity is more pronounced than in any other $\kappa$-type BEDT-TTF or related compounds. The metal-insulator transitions affects the molecular vibrations due to the coupling to the electronic system; in addition we observe a clear splitting of the charge-sensitive vibrational modes below $T_{\rm MI}$ that evidences the presence of two distinct BETS dimers in this compound. The Mn[N(CN)$_2$]$_3^-$ layers are determined by the chain structure of the anions resulting in a rather anisotropic behavior and remarkable temperature dependence of the vibronic features. At low temperatures the ESR properties are affected by the Mn$^{2+}$ ions via $\pi$-$d$-coupling and antiferromagnetic ordering within the $\pi$-spins: The $g$-factor shifts enormously with a pronounced in-plane anisotropy that flips as the temperature decreases; the lines broaden significantly; and the spin susceptibility increases upon cooling with a kink at the phase transition.

Effective communication in automated chat systems hinges on the ability to understand and respond to context. Traditional models often struggle with determining when additional context is necessary for generating appropriate responses. This paper introduces Context-Aware BERT (CA-BERT), a transformer-based model specifically fine-tuned to address this challenge. CA-BERT innovatively applies deep learning techniques to discern context necessity in multi-turn chat interactions, enhancing both the relevance and accuracy of responses. We describe the development of CA-BERT, which adapts the robust architecture of BERT with a novel training regimen focused on a specialized dataset of chat dialogues. The model is evaluated on its ability to classify context necessity, demonstrating superior performance over baseline BERT models in terms of accuracy and efficiency. Furthermore, CA-BERT's implementation showcases significant reductions in training time and resource usage, making it feasible for real-time applications. The results indicate that CA-BERT can effectively enhance the functionality of chatbots by providing a nuanced understanding of context, thereby improving user experience and interaction quality in automated systems. This study not only advances the field of NLP in chat applications but also provides a framework for future research into context-sensitive AI developments., Comment: This paper has been accepted by ICBASE 2024

Horticulture Luminescent Solar Concentrators (HLSCs) represent an innovative concept developed in recent years to promote crop yields, building upon the foundation of traditional Luminescent Solar Concentrators (LSCs). HLSCs are characterized by two distinct properties: spectral conversion and light extraction. Unlike traditional LSCs, HLSCs focus on converting energy from one part of the solar spectrum (typically green) to a specific range (usually red) and aim for the converted photons to exit the device from the bottom surface rather than the edge surfaces. In this study, we start by examining the specific requirements of horticulture to clarify the motivation for using HLSCs. We re-evaluate and propose new optical metrics tailored to HLSCs. Additionally, we analyse potential loss channels for direct red emission and converted red emission. Utilizing Monte Carlo ray tracing method and experimental data, we further explore the factors influencing these loss channels. Our work provides a fundamental discussion on HLSCs and offers design guidelines for future HLSC research.

Recent Text-to-SQL methods leverage large language models (LLMs) by incorporating feedback from the database management system. While these methods effectively address execution errors in SQL queries, they struggle with database mismatches -- errors that do not trigger execution exceptions. Database mismatches include issues such as condition mismatches and stricter constraint mismatches, both of which are more prevalent in real-world scenarios. To address these challenges, we propose a tool-assisted agent framework for SQL inspection and refinement, equipping the LLM-based agent with two specialized tools: a retriever and a detector, designed to diagnose and correct SQL queries with database mismatches. These tools enhance the capability of LLMs to handle real-world queries more effectively. We also introduce Spider-Mismatch, a new dataset specifically constructed to reflect the condition mismatch problems encountered in real-world scenarios. Experimental results demonstrate that our method achieves the highest performance on the averaged results of the Spider and Spider-Realistic datasets in few-shot settings, and it significantly outperforms baseline methods on the more realistic dataset, Spider-Mismatch., Comment: work in progress

Depth estimation from a monocular 360 image is important to the perception of the entire 3D environment. However, the inherent distortion and large field of view (FoV) in 360 images pose great challenges for this task. To this end, existing mainstream solutions typically introduce additional perspective-based 360 representations (\textit{e.g.}, Cubemap) to achieve effective feature extraction. Nevertheless, regardless of the introduced representations, they eventually need to be unified into the equirectangular projection (ERP) format for the subsequent depth estimation, which inevitably reintroduces the troublesome distortions. In this work, we propose an oriented distortion-aware Gabor Fusion framework (PGFuse) to address the above challenges. First, we introduce Gabor filters that analyze texture in the frequency domain, thereby extending the receptive fields and enhancing depth cues. To address the reintroduced distortions, we design a linear latitude-aware distortion representation method to generate customized, distortion-aware Gabor filters (PanoGabor filters). Furthermore, we design a channel-wise and spatial-wise unidirectional fusion module (CS-UFM) that integrates the proposed PanoGabor filters to unify other representations into the ERP format, delivering effective and distortion-free features. Considering the orientation sensitivity of the Gabor transform, we introduce a spherical gradient constraint to stabilize this sensitivity. Experimental results on three popular indoor 360 benchmarks demonstrate the superiority of the proposed PGFuse to existing state-of-the-art solutions. Code can be available upon acceptance.

This paper presents a detailed study of the physical properties of seven C IV absorbers identified at z_abs = 0.68-1.28 along the line of sight toward QSO PG 1522+101 (z_QSO = 1.330). The study leverages high-quality QSO spectra from HST COS and STIS, and Keck HIRES to resolve component structures and to constrain the gas density and elemental abundances of individual components. Under the assumption of photoionization equilibrium (PIE), five of the 12 C IV components require a mixture of high- and low-density phases to fully explain the observed relative abundances between low-, intermediate-, and high-ionization species. In addition, galaxy surveys carried out using VLT MUSE and Magellan LDSS3C are utilized to characterize the galaxy environments. The results of this analysis are summarized as follows: (1) no luminous galaxies (> 0.1 L*) are found within 100 kpc in projected distance from the C IV absorbers; (2) the C IV selection preferentially targets high-metallicity (near solar) and chemically-evolved gas (~ solar [C/O] elemental abundances) in galaxy halos; (3) the observed narrow line widths of individual C IV components, places a stringent limit on the gas temperature (< 5e4 K) and supports a photoionization origin; (4) additional local ionizing sources beyond the UV ionizing background may be necessary for at least one absorber based on the observed deficit of He I relative to H I; and (5) a PIE assumption may not apply when the gas metallicity exceeds the solar value and the component line width implies a warmer temperature than expected from PIE models., Comment: 31 pages, 16 figures, accepted to The Open Journal of Astrophysics with revisions

We present a unified transformer, i.e., Show-o, that unifies multimodal understanding and generation. Unlike fully autoregressive models, Show-o unifies autoregressive and (discrete) diffusion modeling to adaptively handle inputs and outputs of various and mixed modalities. The unified model flexibly supports a wide range of vision-language tasks including visual question-answering, text-to-image generation, text-guided inpainting/extrapolation, and mixed-modality generation. Across various benchmarks, it demonstrates comparable or superior performance to existing individual models with an equivalent or larger number of parameters tailored for understanding or generation. This significantly highlights its potential as a next-generation foundation model. Code and models are released at https://github.com/showlab/Show-o., Comment: Technical Report

The shallow potential wells of star-forming dwarf galaxies make their surrounding circumgalactic and intergalactic medium (CGM/IGM) sensitive laboratories for studying the inflows and outflows thought to regulate galaxy evolution. We present new absorption-line measurements in quasar sightlines probing within projected distances of $<300$ kpc from 91 star-forming field dwarf galaxies with a median stellar mass of $\log{M_\star/\rm{M_\odot}} \approx 8.3$ at $0.077 < z < 0.73$ from the Cosmic Ultraviolet Baryon Survey (CUBS). In this redshift range, the CUBS quasar spectra cover a suite of transitions including H I, low and intermediate metal ions (e.g., C II, Si II, C III, and Si III), and highly ionized O VI. This CUBS-Dwarfs survey enables constraints with samples 9$\times$ larger than past dwarf CGM/IGM studies with similar ionic coverage. We find that low and intermediate ionization metal absorption is rare around dwarf galaxies, consistent with previous surveys of local dwarfs. In contrast, highly ionized O VI is commonly observed in sightlines that pass within the virial radius of a dwarf, and O VI detection rates are non-negligible at projected distances of 1$-$2$\times$ the virial radius. Based on these measurements, we estimate that the O VI-bearing phase of the CGM/IGM accounts for a dominant share of the metal budget of dwarf galaxies. The absorption kinematics suggest that a relatively modest fraction of the O VI-bearing gas is formally unbound. Together, these results imply that low-mass systems at $z\lesssim 1$ effectively retain a substantial fraction of their metals within the nearby CGM and IGM., Comment: Submitted to ApJ. 18 pages, 6 figures, 2 tables

In recent years, Wi-Fi sensing has garnered significant attention due to its numerous benefits, such as privacy protection, low cost, and penetration ability. Extensive research has been conducted in this field, focusing on areas such as gesture recognition, people identification, and fall detection. However, many data-driven methods encounter challenges related to domain shift, where the model fails to perform well in environments different from the training data. One major factor contributing to this issue is the limited availability of Wi-Fi sensing datasets, which makes models learn excessive irrelevant information and over-fit to the training set. Unfortunately, collecting large-scale Wi-Fi sensing datasets across diverse scenarios is a challenging task. To address this problem, we propose CrossFi, a siamese network-based approach that excels in both in-domain scenario and cross-domain scenario, including few-shot, zero-shot scenarios, and even works in few-shot new-class scenario where testing set contains new categories. The core component of CrossFi is a sample-similarity calculation network called CSi-Net, which improves the structure of the siamese network by using an attention mechanism to capture similarity information, instead of simply calculating the distance or cosine similarity. Based on it, we develop an extra Weight-Net that can generate a template for each class, so that our CrossFi can work in different scenarios. Experimental results demonstrate that our CrossFi achieves state-of-the-art performance across various scenarios. In gesture recognition task, our CrossFi achieves an accuracy of 98.17% in in-domain scenario, 91.72% in one-shot cross-domain scenario, 64.81% in zero-shot cross-domain scenario, and 84.75% in one-shot new-class scenario. To facilitate future research, we will release the code for our model upon publication.

Mirkovi\'c introduced the notion of loop Grassmannian for symmetric integer matrix $\kappa$. It is a two-step limit of the local projective space $Z_{\kappa}^{\alpha}$, which generalizes the usual Zastava for a simply laced group $G$. The usual loop Grassmannian of $G$ is recovered when the matrix $\kappa$ is the Cartan matrix of $G$. On the other hand, Braverman, Finkelberg, and Nakajima showed that the Compactified Coulomb branch $\mathbf{M}_{Q}^{\alpha}$ for the quiver gauge theory with no framing also generalizes the usual Zastava. We show that in the case when $\kappa$ is the associated matrix of the quiver $Q$, these two generalizations of Zastava coincide, i.e $\mathbf{M}_{Q}^{\alpha}\cong Z_{\kappa(Q)}^{\alpha}$.

Zero-shot image recognition (ZSIR) aims at empowering models to recognize and reason in unseen domains via learning generalized knowledge from limited data in the seen domain. The gist for ZSIR is to execute element-wise representation and reasoning from the input visual space to the target semantic space, which is a bottom-up modeling paradigm inspired by the process by which humans observe the world, i.e., capturing new concepts by learning and combining the basic components or shared characteristics. In recent years, element-wise learning techniques have seen significant progress in ZSIR as well as widespread application. However, to the best of our knowledge, there remains a lack of a systematic overview of this topic. To enrich the literature and provide a sound basis for its future development, this paper presents a broad review of recent advances in element-wise ZSIR. Concretely, we first attempt to integrate the three basic ZSIR tasks of object recognition, compositional recognition, and foundation model-based open-world recognition into a unified element-wise perspective and provide a detailed taxonomy and analysis of the main research approaches. Then, we collect and summarize some key information and benchmarks, such as detailed technical implementations and common datasets. Finally, we sketch out the wide range of its related applications, discuss vital challenges, and suggest potential future directions., Comment: 23 pages, 7 figures, and 3 tables

Advancements in text-to-image diffusion models have broadened extensive downstream practical applications, but such models often encounter misalignment issues between text and image. Taking the generation of a combination of two disentangled concepts as an example, say given the prompt "a tea cup of iced coke", existing models usually generate a glass cup of iced coke because the iced coke usually co-occurs with the glass cup instead of the tea one during model training. The root of such misalignment is attributed to the confusion in the latent semantic space of text-to-image diffusion models, and hence we refer to the "a tea cup of iced coke" phenomenon as Latent Concept Misalignment (LC-Mis). We leverage large language models (LLMs) to thoroughly investigate the scope of LC-Mis, and develop an automated pipeline for aligning the latent semantics of diffusion models to text prompts. Empirical assessments confirm the effectiveness of our approach, substantially reducing LC-Mis errors and enhancing the robustness and versatility of text-to-image diffusion models. The code and dataset are here: https://github.com/RossoneriZhao/iced_coke., Comment: Accepted by the 18th European Conference on Computer Vision ECCV 2024

The Quick-view (QV) technique serves as a primary method for detecting defects within sewerage systems. However, the effectiveness of QV is impeded by the limited visual range of its hardware, resulting in suboptimal image quality for distant portions of the sewer network. Image super-resolution is an effective way to improve image quality and has been applied in a variety of scenes. However, research on super-resolution for sewer images remains considerably unexplored. In response, this study leverages the inherent depth relationships present within QV images and introduces a novel Depth-guided, Reference-based Super-Resolution framework denoted as DSRNet. It comprises two core components: a depth extraction module and a depth information matching module (DMM). DSRNet utilizes the adjacent frames of the low-resolution image as reference images and helps them recover texture information based on the correlation. By combining these modules, the integration of depth priors significantly enhances both visual quality and performance benchmarks. Besides, in pursuit of computational efficiency and compactness, a super-resolution knowledge distillation model based on an attention mechanism is introduced. This mechanism facilitates the acquisition of feature similarity between a more complex teacher model and a streamlined student model, with the latter being a lightweight version of DSRNet. Experimental results demonstrate that DSRNet significantly improves PSNR and SSIM compared with other methods. This study also conducts experiments on sewer defect semantic segmentation, object detection, and classification on the Pipe dataset and Sewer-ML dataset. Experiments show that the method can improve the performance of low-resolution sewer images in these tasks.

We update the HI surface density measurements for a subset of 17 THINGS galaxies by dealing with the short-spacing problem of the original VLA HI images. It is the same sample that Bigiel et al. (2010) used to study the relation between HI surface densities and star formation rate surface densities in galaxy outer disks, which are beyond the optical radius r25. For ten galaxies, the update is based on combining original THINGS VLA HI images with HI images taken by the single-dish FAST in the FEASTS program. The median increment of HI surface densities in outer disks is 0.15 to 0.4 dex at a given new HI surface density. Several galaxies change significantly in the shape of radial profiles HI surface densities, and seven galaxies are now more than 1-$\sigma$ below the HI size-mass relation. We update the HI star formation laws in outer disks. The median relation between HI surface densities and star formation rate surface densities based on pixelwise measurements shifts downward by around 0.15 dex because the HI surface density values shift rightward, and the scatter increases significantly. The scatter of the relation, indicating the star forming efficiency, exhibits a much stronger positive correlation with the stellar mass surface density than before. Thus, detecting the previously missed, diffuse HI due to short-spacing problem of the VLA observation is important in revealing the true condition and variation of star formation possibly regulated by stellar feedbacks in localized environment of outer disks., Comment: 27 pages, the second paper of "FEASTS Combined with Interferometry". Data is available on the page http://kavli.pku.edu.cn/~jwang/FEASTS_data.html

Automated 3D city generation, focusing on road networks and building layouts, is in high demand for applications in urban design, multimedia games and autonomous driving simulations. The surge of generative AI facilitates designing city layouts based on deep learning models. However, the lack of high-quality datasets and benchmarks hinders the progress of these data-driven methods in generating road networks and building layouts. Furthermore, few studies consider urban characteristics, which generally take graphics as analysis objects and are crucial for practical applications, to control the generative process. To alleviate these problems, we introduce a multimodal dataset with accompanying evaluation metrics for controllable generation of Road networks and Building layouts (RoBus), which is the first and largest open-source dataset in city generation so far. RoBus dataset is formatted as images, graphics and texts, with $72,400$ paired samples that cover around $80,000km^2$ globally. We analyze the RoBus dataset statistically and validate the effectiveness against existing road networks and building layouts generation methods. Additionally, we design new baselines that incorporate urban characteristics, such as road orientation and building density, in the process of generating road networks and building layouts using the RoBus dataset, enhancing the practicality of automated urban design. The RoBus dataset and related codes are published at https://github.com/tourlics/RoBus_Dataset.

Recent years have witnessed a trend that large language model (LLM) based text-to-speech (TTS) emerges into the mainstream due to their high naturalness and zero-shot capacity. In this paradigm, speech signals are discretized into token sequences, which are modeled by an LLM with text as prompts and reconstructed by a token-based vocoder to waveforms. Obviously, speech tokens play a critical role in LLM-based TTS models. Current speech tokens are learned in an unsupervised manner, which lacks explicit semantic information and alignment to the text. In this paper, we propose to represent speech with supervised semantic tokens, which are derived from a multilingual speech recognition model by inserting vector quantization into the encoder. Based on the tokens, we further propose a scalable zero-shot TTS synthesizer, CosyVoice, which consists of an LLM for text-to-token generation and a conditional flow matching model for token-to-speech synthesis. Experimental results show that supervised semantic tokens significantly outperform existing unsupervised tokens in terms of content consistency and speaker similarity for zero-shot voice cloning. Moreover, we find that utilizing large-scale data further improves the synthesis performance, indicating the scalable capacity of CosyVoice. To the best of our knowledge, this is the first attempt to involve supervised speech tokens into TTS models., Comment: work in progress. arXiv admin note: substantial text overlap with arXiv:2407.04051

To efficiently train large-scale models, low-bit gradient communication compresses full-precision gradients on local GPU nodes into low-precision ones for higher gradient synchronization efficiency among GPU nodes. However, it often degrades training quality due to compression information loss. To address this, we propose the Low-bit Communication Adaptor (LoCo), which compensates gradients on local GPU nodes before compression, ensuring efficient synchronization without compromising training quality. Specifically, LoCo designs a moving average of historical compensation errors to stably estimate concurrent compression error and then adopts it to compensate for the concurrent gradient compression, yielding a less lossless compression. This mechanism allows it to be compatible with general optimizers like Adam and sharding strategies like FSDP. Theoretical analysis shows that integrating LoCo into full-precision optimizers like Adam and SGD does not impair their convergence speed on nonconvex problems. Experimental results show that across large-scale model training frameworks like Megatron-LM and PyTorch's FSDP, LoCo significantly improves communication efficiency, e.g., improving Adam's training speed by 14% to 40% without performance degradation on large language models like LLAMAs and MoE.

This report introduces FunAudioLLM, a model family designed to enhance natural voice interactions between humans and large language models (LLMs). At its core are two innovative models: SenseVoice, which handles multilingual speech recognition, emotion recognition, and audio event detection; and CosyVoice, which facilitates natural speech generation with control over multiple languages, timbre, speaking style, and speaker identity. SenseVoice-Small delivers exceptionally low-latency ASR for 5 languages, and SenseVoice-Large supports high-precision ASR for over 50 languages, while CosyVoice excels in multi-lingual voice generation, zero-shot in-context learning, cross-lingual voice cloning, and instruction-following capabilities. The models related to SenseVoice and CosyVoice have been open-sourced on Modelscope and Huggingface, along with the corresponding training, inference, and fine-tuning codes released on GitHub. By integrating these models with LLMs, FunAudioLLM enables applications such as speech-to-speech translation, emotional voice chat, interactive podcasts, and expressive audiobook narration, thereby pushing the boundaries of voice interaction technology. Demos are available at https://fun-audio-llm.github.io, and the code can be accessed at https://github.com/FunAudioLLM., Comment: Work in progress. Authors are listed in alphabetical order by family name

Text-to-video (T2V) generation has recently garnered significant attention thanks to the large multi-modality model Sora. However, T2V generation still faces two important challenges: 1) Lacking a precise open sourced high-quality dataset. The previous popular video datasets, e.g. WebVid-10M and Panda-70M, are either with low quality or too large for most research institutions. Therefore, it is challenging but crucial to collect a precise high-quality text-video pairs for T2V generation. 2) Ignoring to fully utilize textual information. Recent T2V methods have focused on vision transformers, using a simple cross attention module for video generation, which falls short of thoroughly extracting semantic information from text prompt. To address these issues, we introduce OpenVid-1M, a precise high-quality dataset with expressive captions. This open-scenario dataset contains over 1 million text-video pairs, facilitating research on T2V generation. Furthermore, we curate 433K 1080p videos from OpenVid-1M to create OpenVidHD-0.4M, advancing high-definition video generation. Additionally, we propose a novel Multi-modal Video Diffusion Transformer (MVDiT) capable of mining both structure information from visual tokens and semantic information from text tokens. Extensive experiments and ablation studies verify the superiority of OpenVid-1M over previous datasets and the effectiveness of our MVDiT., Comment: 15 pages, 9 figures

Cross-lingual Cross-modal Retrieval (CCR) is an essential task in web search, which aims to break the barriers between modality and language simultaneously and achieves image-text retrieval in the multi-lingual scenario with a single model. In recent years, excellent progress has been made based on cross-lingual cross-modal pre-training; particularly, the methods based on contrastive learning on large-scale data have significantly improved retrieval tasks. However, these methods directly follow the existing pre-training methods in the cross-lingual or cross-modal domain, leading to two problems of inconsistency in CCR: The methods with cross-lingual style suffer from the intra-modal error propagation, resulting in inconsistent recall performance across languages in the whole dataset. The methods with cross-modal style suffer from the inter-modal optimization direction bias, resulting in inconsistent rank across languages within each instance, which cannot be reflected by Recall@K. To solve these problems, we propose a simple but effective 1-to-K contrastive learning method, which treats each language equally and eliminates error propagation and optimization bias. In addition, we propose a new evaluation metric, Mean Rank Variance (MRV), to reflect the rank inconsistency across languages within each instance. Extensive experiments on four CCR datasets show that our method improves both recall rates and MRV with smaller-scale pre-trained data, achieving the new state-of-art., Comment: Accepted by KDD 2024 Research Track

Text embeddings from large language models (LLMs) have achieved excellent results in tasks such as information retrieval, semantic textual similarity, etc. In this work, we show an interesting finding: when feeding a text into the embedding LLMs, the obtained text embedding will be able to be aligned with the key tokens in the input text. We first fully analyze this phenomenon on eight embedding LLMs and show that this phenomenon is universal and is not affected by model architecture, training strategy, and embedding method. With a deeper analysis, we then find that the main change in embedding space between the embedding LLMs and their original generative LLMs is in the first principal component. By adjusting the first principal component, we can align text embedding with the key tokens. Finally, we give several examples to demonstrate the vast application potential of this finding: (1) we propose a simple and practical sparse retrieval method based on the aligned tokens, which can achieve 80\% of the dense retrieval effect of the same model while reducing the computation significantly; (2) we show that our findings provide a fresh perspective to help understand fuzzy concepts (e.g., semantic relatedness vs. semantic similarity) and emerging technologies (e.g., instruction-following embedding) in this field., Comment: Work in Progress

Diffusion models (DMs) have achieved significant success in generating imaginative images given textual descriptions. However, they are likely to fall short when it comes to real-life scenarios with intricate details. The low-quality, unrealistic human faces in text-to-image generation are one of the most prominent issues, hindering the wide application of DMs in practice. Targeting addressing such an issue, we first assess the face quality of generations from popular pre-trained DMs with the aid of human annotators and then evaluate the alignment between existing metrics with human judgments. Observing that existing metrics can be unsatisfactory for quantifying face quality, we develop a novel metric named FaceScore (FS) by fine-tuning the widely used ImageReward on a dataset of (win, loss) face pairs cheaply crafted by an inpainting pipeline of DMs. Extensive studies reveal FS enjoys a superior alignment with humans. On the other hand, FS opens up the door for enhancing DMs for better face generation. With FS offering image ratings, we can easily perform preference learning algorithms to refine DMs like SDXL. Comprehensive experiments verify the efficacy of our approach for improving face quality. The code is released at https://github.com/OPPO-Mente-Lab/FaceScore., Comment: Under review

Particle methods are less computationally efficient than grid based numerical solution of the Navier Stokes equation. However, they have important advantages including rigorous mass conservation, momentum conservation and isotropy. In addition, there is no need for explicit interface tracking/capturing and code development effort is relatively low. We describe applications of three particle methods: molecular dynamics, dissipative particle dynamics and smoothed particle hydrodynamics. The mesoscale (between the molecular and continuum scales) dissipative particle dynamics method can be used to simulate systems that are too large to simulate using molecular dynamics but small enough for thermal fluctuations to play an important role., Comment: 10 pages, 10 figures

Symmetric submodular maximization is an important class of combinatorial optimization problems, including MAX-CUT on graphs and hyper-graphs. The state-of-the-art algorithm for the problem over general constraints has an approximation ratio of $0.432$. The algorithm applies the canonical continuous greedy technique that involves a sampling process. It, therefore, suffers from high query complexity and is inherently randomized. In this paper, we present several efficient deterministic algorithms for maximizing a symmetric submodular function under various constraints. Specifically, for the cardinality constraint, we design a deterministic algorithm that attains a $0.432$ ratio and uses $O(kn)$ queries. Previously, the best deterministic algorithm attains a $0.385-\epsilon$ ratio and uses $O\left(kn (\frac{10}{9\epsilon})^{\frac{20}{9\epsilon}-1}\right)$ queries. For the matroid constraint, we design a deterministic algorithm that attains a $1/3-\epsilon$ ratio and uses $O(kn\log \epsilon^{-1})$ queries. Previously, the best deterministic algorithm can also attain $1/3-\epsilon$ ratio but it uses much larger $O(\epsilon^{-1}n^4)$ queries. For the packing constraints with a large width, we design a deterministic algorithm that attains a $0.432-\epsilon$ ratio and uses $O(n^2)$ queries. To the best of our knowledge, there is no deterministic algorithm for the constraint previously. The last algorithm can be adapted to attain a $0.432$ ratio for single knapsack constraint using $O(n^4)$ queries. Previously, the best deterministic algorithm attains a $0.316-\epsilon$ ratio and uses $\widetilde{O}(n^3)$ queries.

Reducing the inference latency of large language models (LLMs) is crucial, and speculative decoding (SD) stands out as one of the most effective techniques. Rather than letting the LLM generate all tokens directly, speculative decoding employs effective proxies to predict potential outputs, which are then verified by the LLM without compromising the generation quality. Yet, deploying SD in real online LLM serving systems (with continuous batching) does not always yield improvement -- under higher request rates or low speculation accuracy, it paradoxically increases latency. Furthermore, there is no best speculation length work for all workloads under different system loads. Based on the observations, we develop a dynamic framework SmartSpec. SmartSpec dynamically determines the best speculation length for each request (from 0, i.e., no speculation, to many tokens) -- hence the associated speculative execution costs -- based on a new metric called goodput, which characterizes the current observed load of the entire system and the speculation accuracy. We show that SmartSpec consistently reduces average request latency by up to 3.2x compared to non-speculative decoding baselines across different sizes of target models, draft models, request rates, and datasets. Moreover, SmartSpec can be applied to different styles of speculative decoding, including traditional, model-based approaches as well as model-free methods like prompt lookup and tree-style decoding.

Mixture of experts (MoE) has become the standard for constructing production-level large language models (LLMs) due to its promise to boost model capacity without causing significant overheads. Nevertheless, existing MoE methods usually enforce a constant top-k routing for all tokens, which is arguably restrictive because various tokens (e.g., "" vs. "apple") may require various numbers of experts for feature abstraction. Lifting such a constraint can help make the most of limited resources and unleash the potential of the model for downstream tasks. In this sense, we introduce AdaMoE to realize token-adaptive routing for MoE, where different tokens are permitted to select a various number of experts. AdaMoE makes minimal modifications to the vanilla MoE with top-k routing -- it simply introduces a fixed number of null experts, which do not consume any FLOPs, to the expert set and increases the value of k. AdaMoE does not force each token to occupy a fixed number of null experts but ensures the average usage of the null experts with a load-balancing loss, leading to an adaptive number of null/true experts used by each token. AdaMoE exhibits a strong resemblance to MoEs with expert choice routing while allowing for trivial auto-regressive modeling. AdaMoE is easy to implement and can be effectively applied to pre-trained (MoE-)LLMs. Extensive studies show that AdaMoE can reduce average expert load (FLOPs) while achieving superior performance. For example, on the ARC-C dataset, applying our method to fine-tuning Mixtral-8x7B can reduce FLOPs by 14.5% while increasing accuracy by 1.69%., Comment: Findings of EMNLP 2024

The wave function Ansatze are crucial in the context of the Variational Quantum Eigensolver (VQE). In the Noisy Intermediate-Scale Quantum era, the design of low-depth wave function Ansatze is of great importance for executing quantum simulations of electronic structure on noisy quantum devices. In this work, we present a circuit-efficient implementation of two-body Qubit-Excitation-Based (QEB) operator for building shallow-circuit wave function Ansatze within the framework of Adaptive Derivative-Assembled Pseudo-Trotter (ADAPT) VQE. This new algorithm is applied to study ground- and excited-sate problems for small molecules, demonstrating significant reduction of circuit depths compared to fermionic excitation-based and QEB ADAPT-VQE algorithms. This circuit-efficient algorithm shows great promise for quantum simulations of electronic structures, leading to improved performance on current quantum hardware.

Large Language Models (LLMs) have shown great potential in code generation. However, current LLMs still cannot reliably generate correct code. Moreover, it is unclear what kinds of code generation errors LLMs can make. To address this, we conducted an empirical study to analyze incorrect code snippets generated by six popular LLMs on the HumanEval dataset. We analyzed these errors alongside two dimensions of error characteristics -- semantic characteristics and syntactic characteristics -- to derive a comprehensive code generation error taxonomy for LLMs through open coding and thematic analysis. We then labeled all 557 incorrect code snippets based on this taxonomy. Our results showed that the six LLMs exhibited similar distributions of syntactic characteristics while different distributions of semantic characteristics. Furthermore, we analyzed the correlation between different error characteristics and factors such as task complexity, code length, and test-pass rate. Finally, we highlight the challenges that LLMs may encounter when generating code and propose implications for future research on reliable code generation with LLMs., Comment: Extended from our MAPS 2023 paper. Our data is available at https://llm-code-errors.cs.purdue.edu