Author name cluster

Tian Wang

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

8 papers

2 author rows

ICLR Conference 2025 Conference Paper

OmniKV: Dynamic Context Selection for Efficient Long-Context LLMs

Jitai Hao
Yuke Zhu
Tian Wang
Jun Yu 0002
Xin Xin 0003
Bo Zheng 0007
Zhaochun Ren
Sheng Guo 0005

During the inference phase of Large Language Models (LLMs) with long context, a substantial portion of GPU memory is allocated to the KV cache, with memory usage increasing as the sequence length grows. To mitigate the GPU memory footprint associate with KV cache, some previous studies have discarded less important tokens based on the sparsity identified in attention scores in long context scenarios. However, we argue that attention scores cannot indicate the future importance of tokens in subsequent generation iterations, because attention scores are calculated based on current hidden states. Therefore, we propose OmniKV, a token-dropping-free and training-free inference method, which achieves a 1.68x speedup without any loss in performance. It is well-suited for offloading, significantly reducing KV cache memory usage by up to 75% with it. The core innovative insight of OmniKV is: Within a single generation iteration, there is a high degree of similarity in the important tokens identified across consecutive layers. Extensive experiments demonstrate that OmniKV achieves state-of-the-art performance across multiple benchmarks, with particularly advantages in chain-of-thoughts scenarios. OmniKV extends the maximum context length supported by a single A100 for Llama-3-8B from 128K to 450K. Our code is available at https://github.com/antgroup/OmniKV.git.

Details

TMLR Journal 2025 Journal Article

Rec-R1: Bridging Generative Large Language Models and User-Centric Recommendation Systems via Reinforcement Learning

Jiacheng Lin
Tian Wang
Kun Qian

We propose Rec-R1, a general reinforcement learning framework that bridges large language models (LLMs) with recommendation systems through closed-loop optimization. Unlike prompting and supervised fine-tuning (SFT), Rec-R1 directly optimizes LLM generation using feedback from a fixed, black-box recommendation model—without relying on synthetic SFT data from proprietary models like GPT-4o. This avoids the substantial cost and effort required for data distillation. To verify the effectiveness of Rec-R1, we evaluate Rec-R1 on three representative tasks: product search, sequential recommendation, and product re-ranking. Experimental results demonstrate that Rec-R1 not only consistently outperforms prompting- and SFT-based methods, but also achieves remarkable gains over strong discriminative baselines, even when used with simple retrievers like BM25. More impressively, Rec-R1 preserves the general-purpose capabilities of the LLM, in contrast to SFT, which often impairs instruction-following and reasoning. These findings suggest Rec-R1 as a promising foundation for continual task-specific adaptation without catastrophic forgetting.

PDF Details

NeurIPS Conference 2024 Conference Paper

Latent Neural Operator for Solving Forward and Inverse PDE Problems

Tian Wang
Chuang Wang

Neural operators effectively solve PDE problems from data without knowing the explicit equations, which learn the map from the input sequences of observed samples to the predicted values. Most existing works build the model in the original geometric space, leading to high computational costs when the number of sample points is large. We present the Latent Neural Operator (LNO) solving PDEs in the latent space. In particular, we first propose Physics-Cross-Attention (PhCA) transforming representation from the geometric space to the latent space, then learn the operator in the latent space, and finally recover the real-world geometric space via the inverse PhCA map. Our model retains flexibility that can decode values in any position not limited to locations defined in the training set, and therefore can naturally perform interpolation and extrapolation tasks particularly useful for inverse problems. Moreover, the proposed LNO improves both prediction accuracy and computational efficiency. Experiments show that LNO reduces the GPU memory by 50%, speeds up training 1. 8 times, and reaches state-of-the-art accuracy on four out of six benchmarks for forward problems and a benchmark for inverse problem. Code is available at https: //github. com/L-I-M-I-T/LatentNeuralOperator.

PDF Details DOI

IJCAI Conference 2023 Conference Paper

Learning Attention from Attention: Efficient Self-Refinement Transformer for Face Super-Resolution

Guanxin Li
Jingang Shi
Yuan Zong
Fei Wang
Tian Wang
Yihong Gong

Recently, Transformer-based architecture has been introduced into face super-resolution task due to its advantage in capturing long-range dependencies. However, these approaches tend to integrate global information in a large searching region, which neglect to focus on the most relevant information and induce blurry effect by the irrelevant textures. Some improved methods simply constrain self-attention in a local window to suppress the useless information. But it also limits the capability of recovering high-frequency details when flat areas dominate the local searching window. To improve the above issues, we propose a novel self-refinement mechanism which could adaptively achieve texture-aware reconstruction in a coarse-to-fine procedure. Generally, the primary self-attention is first conducted to reconstruct the coarse-grained textures and detect the fine-grained regions required further compensation. Then, region selection attention is performed to refine the textures on these key regions. Since self-attention considers the channel information on tokens equally, we employ a dual-branch feature integration module to privilege the important channels in feature extraction. Furthermore, we design the wavelet fusion module which integrate shallow-layer structure and deep-layer detailed feature to recover realistic face images in frequency domain. Extensive experiments demonstrate the effectiveness on a variety of datasets.

PDF Details DOI

TCS Journal 2023 Journal Article

Theoretical design of decentralized auction framework under mobile crowdsourcing environment

Jianxiong Guo
Xingjian Ding
Tian Wang
Weijia Jia

With the rapid popularization of mobile devices, the mobile crowdsourcing has become a hot topic in order to make full use of the resources of mobile devices. To achieve this goal, it is necessary to design an excellent incentive mechanism to encourage more mobile users to actively undertake crowdsourcing tasks, so as to achieve maximization of certain economic indicators. However, most of the reported incentive mechanisms in the existing literature adopt a centralized platform, which collects the bidding information from workers and task requesters. There is a risk of privacy exposure. In this paper, we design a decentralized auction framework where mobile workers are sellers and task requesters are buyers. This requires each participant to make its own local and independent decision, thereby avoiding centralized processing of task allocation and pricing. Both of them aim to maximize their utilities under the budget constraint. We theoretically prove that our proposed framework is individual rational, budget balanced, truthful, and computationally efficient, and then we conduct a group of numerical simulations to demonstrate its correctness and effectiveness.

Details DOI

EAAI Journal 2023 Journal Article

Uncertainty quantification in molecular property prediction through spherical mixture density networks

Wentao Fan
Lidan Zeng
Tian Wang

As uncertainty quantification is crucial for determining undesirable inputs and improving decisions made by a system to acquire accurate evaluations, it has received much attention in recent years. Motivated by the fact that probability is one of the most effective ways to estimate uncertainty, in this work we propose an effective probabilistic model for quantifying predictive uncertainty in the task of predicting chemical molecular properties. Our model is formulated by developing a spherical mixture density network that is composed of von Mises-Fisher kernel distributions to model graph-structured molecule representations. Furthermore, an ensemble framework for spherical mixture density networks is developed, which can yield high quality predictive uncertainty estimates and obtain better confidence intervals reflecting the sources of these uncertainties in predictions. The effectiveness of our approach in modeling the output predictive uncertainty is validated through empirical analysis on molecular property prediction tasks with two publicly available chemical molecule data sets. Comparing with the current state-of-the-art baselines, our model can better model predictive uncertainty in terms of higher log-likelihood of the data, and reveal that there might be more than one acceptable chemical property associated with an input molecule representation.

Details DOI

TIST Journal 2019 Journal Article

Crowdsourcing Mechanism for Trust Evaluation in CPCS Based on Intelligent Mobile Edge Computing

Tian Wang
Hao Luo
Xi Zheng
Mande Xie

Both academia and industry have directed tremendous interest toward the combination of Cyber Physical Systems and Cloud Computing, which enables a new breed of applications and services. However, due to the relative long distance between remote cloud and end nodes, Cloud Computing cannot provide effective and direct management for end nodes, which leads to security vulnerabilities. In this article, we first propose a novel trust evaluation mechanism using crowdsourcing and Intelligent Mobile Edge Computing. The mobile edge users with relatively strong computation and storage ability are exploited to provide direct management for end nodes. Through close access to end nodes, mobile edge users can obtain various information of the end nodes and determine whether the node is trustworthy. Then, two incentive mechanisms, i.e., Trustworthy Incentive and Quality-Aware Trustworthy Incentive Mechanisms, are proposed for motivating mobile edge users to conduct trust evaluation. The first one aims to motivate edge users to upload their real information about their capability and costs. The purpose of the second one is to motivate edge users to make trustworthy effort to conduct tasks and report results. Detailed theoretical analysis demonstrates the validity of Quality-Aware Trustworthy Incentive Mechanism from data trustfulness, effort trustfulness, and quality trustfulness, respectively. Extensive experiments are carried out to validate the proposed trust evaluation and incentive mechanisms. The results corroborate that the proposed mechanisms can efficiently stimulate mobile edge users to perform evaluation task and improve the accuracy of trust evaluation.

Details DOI

TAAS Journal 2017 Journal Article

e-Sampling

Md Zakirul Alam Bhuiyan
Jie Wu
Guojun Wang
Tian Wang
Mohammad Mehedi Hassan

Sampling rate adaptation is a critical issue in many resource-constrained networked systems, including Wireless Sensor Networks (WSNs). Existing algorithms are primarily employed to detect events such as objects or physical changes at a high, low, or fixed frequency sampling usually adapted by a central unit or a sink, therefore requiring additional resource usage. Additionally, this algorithm potentially makes a network unable to capture a dynamic change or event of interest, which therefore affects monitoring quality. This article studies the problem of a fully autonomous adaptive sampling regarding the presence of a change or event. We propose a novel scheme, termed “event-sensitive adaptive sampling and low-cost monitoring (e-Sampling)” by addressing the problem in two stages, which leads to reduced resource usage (e.g., energy, radio bandwidth). First, e-Sampling provides the embedded algorithm to adaptive sampling that automatically switches between high- and low-frequency intervals to reduce the resource usage, while minimizing false negative detections. Second, by analyzing the frequency content, e-Sampling presents an event identification algorithm suitable for decentralized computing in resource-constrained networks. In the absence of an event, the “uninteresting” data is not transmitted to the sink. Thus, the energy cost is further reduced. e-Sampling can be useful in a broad range of applications. We apply e-Sampling to Structural Health Monitoring (SHM) and Fire Event Monitoring (FEM), which are typical applications of high-frequency events. Evaluation via both simulations and experiments validates the advantages of e-Sampling in low-cost event monitoring, and in effectively expanding the capacity of WSNs for high data rate applications.

Details DOI