Shiqi Liu

I am currently pursuing a Ph.D. in Intelligent Science and Technology at Huazhong University of Science and Technology, advised by Prof. Yihua Tan. I received the Master's Degree in Applied Mathematics from Xi'an Jiaotong University, advised by Prof. Deyu Meng.

Prior to that, I received the Bachelor’s Degree in Math and Applied Math(Honor Science Program and Special Class for Gifted Young) from Xi’an Jiaotong University.

Current Research Interests

My research focuses on how to enable agents to autonomously generate emotions, allowing them to pursue high-level goals—such as survival, human well-being, social connectedness, happiness, a sense of control, and self-identity—while discovering and generating goals and dynamically regulating the priorities of lower-level goals, including task completion, exploration, energy homeostasis, social homeostasis, and safety. Ultimately, I aim to develop a foundation for agents and embodied intelligence to integrate into human life and achieve long-term autonomous operation in complex and changing environments, particularly when faced with multiple autonomous goals that may compete or even conflict with one another (e.g., self-preservation versus promoting human well-being).

Emotion Theory

My work is grounded in a goal-directed theory of emotion, which posits that the core mechanism underlying emotion generation is a goal-directed control cycle. Each cycle involves detecting discrepancies between the current state and a desired goal state, followed by selecting strategies or actions that can reduce those discrepancies. Multiple goal-directed cycles may operate simultaneously and compete for control of behavior. From this perspective, the fundamental distinction between emotional and non-emotional decision-making lies not in the underlying decision-making mechanism itself, but in the value of the goals involved. Emotional responses arise when goals that are particularly important to the individual are activated, threatened, or substantially obstructed.

Past Research Directions

My previous research has also spanned a range of topics, including multimodal large language models, interpretability, generative models, self-paced learning, Bayesian learning, generalization, remote sensing, and image fusion.

News

Jul 1, 2020	I would like to express my sincere gratitude to the School of Mathematics and Statistics of Xi’an Jiaotong University for bestowing upon me the Excellent Master Thesis Award.
Jul 1, 2017	I would like to express my sincere gratitude to the School of Mathematics and Statistics of Xi’an Jiaotong University for bestowing upon me the Excellent Bachelor Thesis Award.

Selected Publications

arXiv
Emotional Preferences as Goal-Priority Regulation

Shiqi Liu, Yihua Tan, Hu Fu, and Guanyu Qi

arXiv preprint arXiv:2608.27072 (arXiv) , 2026

PDF Abs Bib Code

A core question in autonomous decision-making for artificial agents is whether the relative priorities of competing lower-level objectives can be determined by emotional preferences autonomously generated by higher-level goals, rather than being externally prespecified. Under changing external environments and evolving internal states, emotions play an important functional role in regulating the relative priorities of competing goals. Inspired by the goal-directed theory of emotion, this paper studies how such preference regulation can be computationally realized through reinforcement learning. To this end, we first propose a conception of emergent emotional preference: a high-level goal autonomously induces state-dependent preferences over competing lower-level objectives. This conception is built upon a framework consisting of a pretrained multi-objective reinforcement learning (MORL) inner controller and an outer preference generator. The inner controller provides a repertoire of preference-conditioned goal-directed behaviors, while the outer preference generator learns a mapping from the current state to objective preferences through reinforcement learning on a high-level goal. We operationalize emotional preference as a state-dependent regulation of relative goal priorities that emerges through optimization, rather than as predefined emotion labels or a complete model of human emotions. Furthermore, we characterize the policy space induced by preference regulation and derive an upper bound on the optimality gap in terms of the representation error of the inner behavioral repertoire. We show that the gap vanishes when the optimal policy can be represented by the available preference-conditioned policies, meaning that the policy space contains the optimal policy at the representational level. Experiments in self-constructed basic and advanced multi-objective exploration environments (Grid Fruit Tree Battery Exploration) show that the learned preference function exhibits contextual priority switching, graded trade-offs, and temporal persistence, and outperforms the evaluated fixed-preference and handcrafted-preference strategies. These results reveal a computational mechanism by which high-level goal optimization can induce state-dependent emotional preference and dynamically reorganize competing goal-directed behaviors.
@article{liu2026emotional, title={Emotional Preferences as Goal-Priority Regulation}, author={Liu, Shiqi and Tan, Yihua and Fu, Hu and Qi, Guanyu}, journal={arXiv preprint arXiv:2608.27072}, year={2026} }
arXiv
When Does the Concept of "Dog" Emerge in an Audio LLM?

Zhe Wang, Shiqi Liu, Ruiyun Zhong, Tiechong Zhu, and Yihua Tan

arXiv preprint arXiv:2609.33458 (arXiv) , 2026

PDF Abs Bib

Multimodal large language models answer audio questions, but how they represent auditory semantics and use them in decisions remains unclear, limiting our understanding of response formation. We study dog barking in Qwen2.5-Omni-7B using Jacobian lens (J-lens) readout and directional interventions. We define the dog direction as a J-lens-derived hidden-state vector associated with dog; adding or removing its component modulates dog-related information. We find this information decodable without dog/bark prompt cues or animal-identification requirements. Directional interventions change response tendencies and some final answers, with effects concentrated in late-layer states immediately before generation across species classification, vocalization classification, and sound description. The dog direction shows no comparable advantage over controls in animal/other classification. These results provide causal-intervention evidence that the dog direction affects output scores in a task-dependent manner, most consistently at L22 and L24 immediately before generation.
@article{wang2026does, title={When Does the Concept of "Dog" Emerge in an Audio LLM?}, author={Wang, Zhe and Liu, Shiqi and Zhong, Ruiyun and Zhu, Tiechong and Tan, Yihua}, journal={arXiv preprint arXiv:2609.33458}, year={2026} }
arXiv
Understanding Generalization via Set Theory

Shiqi Liu

arXiv preprint arXiv:2311.06545 (arXiv) , 2023

PDF Abs Bib

Generalization is at the core of machine learning models. However, the definition of generalization is not entirely clear. We employ set theory to introduce the concepts of algorithms, hypotheses, and dataset generalization. We analyze the properties of dataset generalization and prove a theorem on surrogate generalization procedures. This theorem leads to our generalization method. Through a generalization experiment on the MNIST dataset, we obtain 13,541 sample bases. When we use the entire training set to evaluate the model's performance, the models achieve an accuracy of 99.945%. However, if we shift the sample bases or modify the neural network structure, the performance experiences a significant decline. We also identify consistently mispredicted samples and find that they are all challenging examples. The experiments substantiated the accuracy of the generalization definition and the effectiveness of the proposed methods. Both the set-theoretic deduction and the experiments help us better understand generalization.
@article{liu2023understanding, title={Understanding Generalization via Set Theory}, author={Liu, Shiqi}, journal={arXiv preprint arXiv:2311.06545}, year={2023} }
Master
深度表示学习在时序、图像与视频数据上的应用初探(Primary Applications of Deep Representation Learning in Time Series, Image and Video Data)

Shiqi Liu

Xi’an jiaotong University Master Thesis (Master) , 2020

PDF Abs Bib

The advent of the era of big data has brought about a drastic change in the basic model of machine learning. The traditional method of artificial feature extraction has been unable to meet the actual needs of diversified, complicated, and quantified application goals. Deep learning has gradually become the mainstream of applications due to its end-to-end data-driven adaptive feature extraction, which has attracted widespread attention in academia and industry. Based on the current research status, this article introduces representative deep learning methods for typical data types such as time series, images, and videos. Based on a deep understanding of the characteristics of the corresponding deep learning methods, it is applied to application problems such as hydrological time series processing, image factor learning, and emotional prediction of EEG data. The paper includes the following research contents. First of all, we use the deep long short-term memory network (LSTM) to predict the daily runoff data of hydrological stations. Using the long short-term memory characteristics of LSTM, the deep LSTM achieves good results in Yichang Hydrological Station. We further compare the relative systematic error, relative standard deviation and relative error range of prediction results of LSTM and backpropagation neural network(BPNN) in flood season of Yi Chang, and the comparison results verify the superiority of the deep LSTM method over BPNN in predicting daily runoff in flood season. In order to verify the effectiveness of the proposed method, we compared the results of LSTM models and deep LSTM models on the data obtained from Cuntan, Wanxian, Fengjie and Yichang hydrological stations. The results show that both moels are suitable for daily runoff data, and deep LSTM models achieve better prediction results. The second part is about one of the main models of deep representation learning, the variational autoencoder (VAE). We focus on supervising the factors extracted by VAE. VAE is used to learn the independent low-dimensional representation of images, but it faces the problem that some pre-specified factors are ignored. We assert that the mutual information of the input and each learned factor of the representation plays a necessary indicator to discover the influential factors. At the same time, we delve into the objective function of VAE, which shows that it tends to induce the sparsity of mutual information of factors when it exceeds the essential dimension of the data, resulting in some non-influential factors which can be ignored and which have negligible data reconstruction capabilities. We show that mutual information can affect the lower bound of VAE reconstruction errors and down-stream classification tasks. We propose an algorithm to calculate the mutual information indicator of VAE and prove its consistency. Experimental results on the MNIST, CelebA, and DEAP datasets show that mutual information helps determine influential generating factors. Besides some generation factors are interpretable, which can help down-stream generation and classification tasks. At last, we focus on deep representation learning applied to EEG video data emotion recognition. By characterizing the VAE and LSTM-based EEG emotion representation learning architecture, the method reaches the latest level. At the same time, the model can also monitor the learned feature shapes. Summarizing the contents of the above three parts, the deep representation learning achieves good performance in time series, image, and video data and can reach the international frontier in some known application fields, such as hydrological prediction, image factor learning, and EEG emotion recognition. Therefore, this paper has significant research and application inspiration for the methodological expansion of these application fields and the methodological level of deep learning itself.
@articleliu2020Survey, title=自步学习研究综述(A Survey On Self-paced Learning), author=Liu, Shiqi, journal=Xi’an jiaotong University Bachelor, year=2020, publisher=
Bachelor
自步学习研究综述(A Survey On Self-paced Learning)

Shiqi Liu

Xi’an jiaotong University Bachelor Thesis (Bachelor) , 2017

PDF Abs Bib Slides

In the era of big data, a lot of information is contained in a large amount of imperfect data. In order to extract useful information from those data, machine learning algorithms should be provided with the abilities of both data selection and data learning. To solve this problem, drawing lessons from the principle of human curriculum learning, which is to learn from simple to difficult adaptively, a new learning regime called self-paced learning(SPL) has appeared in recent years. SPL has developed rapidly and made extensive applications in the fields of machine learning, mainly including semi-supervised learning, weakly supervised learning, robust learning, and very few samples learning. This paper surveys recent achievements and related applications of SPL in machine learning fields. We review the classical curriculum learning and SPL regimes, further, summarize the internal prior knowledge utilized by the SPL and external curriculum prior knowledge utilized by the self-paced curriculum learning, SPL with diversity and self-paced co-training. Current advancements of optimization theory regarding SPL are introduced and united together under the concave conjugacy of SPL. In particular, we first raise the probability viewpoint for SPL, obtain the upper bound of the optimal performance as well as the influence factor of SPL and render a general probability framework for SPL. In the end, we look into the future of the SPL in the closed and open learning environment. The general probability viewpoint could motivate the design of the easiness standard, the formulation of novel data selection principles, the integration with other learning algorithms regarding the self-paced methodologies which would further benefit the promotion and applications of SPL in practical problems.
@articleliu2017Survey, title=自步学习研究综述(A Survey On Self-paced Learning), author=Liu, Shiqi, journal=Xi’an jiaotong University Bachelor, year=2017, publisher=
Pattern Recognition
Discovering influential factors in variational autoencoders

Shiqi Liu, Jingxin Liu, Qian Zhao, Xiangyong Cao, Huibin Li, Deyu Meng, Hongying Meng, and Sheng Liu

Pattern Recognition (Pattern Recognition) , 2020

PDF Abs Bib Code

In the field of machine learning, it is still a critical issue to identify and supervise the learned representation without manually intervening or intuition assistance to extract useful knowledge or serve for the downstream tasks. In this work, we focus on supervising the influential factors extracted by the variational autoencoder (VAE). The VAE is proposed to learn independent low dimension representation while facing the problem that sometimes pre-set factors are ignored. We argue that the mutual information of the input and each learned factor of the representation plays a necessary indicator of discovering the influential factors. We find the VAE objective inclines to induce mutual information sparsity in factor dimension over the data intrinsic dimension and therefore result in some non-influential factors whose function on data reconstruction could be ignored. We show mutual information also influences the lower bound of VAE’s reconstruction error and downstream classification task. To make such indicator applicable, we design an algorithm for calculating the mutual information for VAE and prove its consistency. Experimental results on MNIST, CelebA and DEAP datasets show that mutual information can help determine influential factors, of which some are interpretable and can be used to further generation and classification tasks, and help discover the variant that connects with emotion on DEAP dataset.
@articleliu2020Discovering, title=Discovering influential factors in variational autoencoders, author=Liu, Shiqi and Liu, Jingxin and Zhao, Qian and Cao, Xiangyong and Li, Huibin and Meng, Deyu and Meng, Hongying and Liu, Sheng, journal=Pattern Recognition, year=2020, publisher=Elsevier
Openreview
Preliminary theoretical troubleshooting in variational autoencoder

Shiqi Liu, Qian Zhao, Xiangyong Cao, and Deyu Meng

Openreview (Openreview) , 2018

PDF Abs Bib Code

What would be learned by variational autoencoder(VAE) and what influence the disentanglement of VAE? This paper tries to preliminarily address VAE’s intrinsic dimension, real factor, disentanglement and indicator issues theoretically in the idealistic situation and implementation issue practically through noise modeling perspective in the realistic case. On intrinsic dimension issue, due to information conservation, the idealistic VAE learns and only learns intrinsic factor dimension. Besides, suggested by mutual information separation property, the constraint induced by Gaussian prior to the VAE objective encourages the information sparsity in dimension. On disentanglement issue, subsequently, inspired by information conservation theorem the clarification on disentanglement in this paper is made. On real factor issue, due to factor equivalence, the idealistic VAE possibly learns any factor set in the equivalence class. On indicator issue, the behavior of current disentanglement metric is discussed, and several performance indicators regarding the disentanglement and generating influence are subsequently raised to evaluate the performance of VAE model and to supervise the used factors. On implementation issue, the experiments under noise modeling and constraints empirically testify the theoretical analysis and also show their own characteristic in pursuing disentanglement.
@articleliu2018preliminary, title=Preliminary theoretical troubleshooting in variational autoencoder, author=Liu, Shiqi and Zhao, Qian, and Cao, Xiangyong and Meng, Deyu, journal=Openreview, year=2018, publisher=Openreview
Openreview
Fitting Data Noise in Variational Autoencoder

Shiqi Liu, Qian Zhao, Xiangyong Cao, and Deyu Meng

Openreview (Openreview) , 2017

PDF Abs Bib Code

Why does variational autoencoder(VAE) suffer from bad reconstruction, and what influence the disentanglement of VAE? This paper tries to address those issues through a noise modelling perspective. On one fold, the paper proposes the adaptive noise learning algorithms of Gaussian noise and mixture Gaussian noise assumption which empirically contributes to a better reconstruction than original VAE noise assumptions. On other fold, several generating factor properties in the idealistic VAE case are discussed and several performance indicators regarding the disentanglement and generating influence are subsequently raised to evaluate the performance of VAE model and to supervise the used factors. Theoretical analysis is reflexed in the experiment results.
@articleliu2017fitting, title=Fitting Data Noise in Variational Autoencoder, author=Liu, Shiqi and Zhao, Qian, and Cao, Xiangyong and Meng, Deyu, journal=Openreview, year=2017, publisher=Openreview
CIS
Understanding self-paced learning under concave conjugacy theory

Shiqi Liu, Zilu Ma, Deyu Meng, Kai-Dong Wang, and Yong Zhang

Communiations in Information and Systems (CIS) , 2018

PDF Abs Bib Slides

By simulating the easy-to-hard learning manners of humans/animals, the learning regimes called curriculum learning (CL) and selfpaced learning (SPL) have been recently investigated and invoked broad interests. However, the intrinsic mechanism for analyzing why such learning regimes can work has not been comprehensively investigated. To this issue, this paper proposes a concave conjugacy theory for looking into the insight of CL/SPL. Specifically, by using this theory, we prove the equivalence of the SPL regime and a latent concave objective, which is closely related to the known non-convex regularized penalty widely used in statistics and machine learning. Beyond the previous theory for explaining CL/SPL insights, this new theoretical framework on one hand facilitates two direct approaches for designing new SPL models for certain tasks, and on the other hand can help conduct the latent objective of selfpaced curriculum learning, which is the advanced version of both CL/SPL and possess advantages of both learning regimes to a certain extent. This further facilitates a theoretical understanding for SPCL, instead of only CL/SPL as conventional. Under this theory, we attempt to attain intrinsic latent objectives of two curriculum forms, the partial order and group curriculums, which easily follow the theoretical understanding of the corresponding SPCL regimes.
@articleliu2018understanding, title=Understanding self-paced learning under concave conjugacy theory, author=Liu, Shiqi and Ma, Zilu and Meng, Deyu and Wang, Kai-Dong and Zhang, Yong, journal=Communiations in Information and Systems, year=2018, publisher=intlpress
Information Sciences
On convergence properties of implicit self-paced objective

Zilu Ma, Shiqi Liu, Deyu Meng, Yong Zhang, SioLong Lo, and Zhi Han

Information Sciences (Information Sciences) , 2018

PDF Abs Bib

Self-paced learning (SPL) is a new methodology that simulates the learning principle of humans/animals to start learning easier aspects of a learning task, and then gradually take more complex examples into training. This new-coming learning regime has been empirically substantiated to be effective in various computer vision and pattern recognition tasks. Recently, it has been proved that the SPL regime has a close relationship to a implicit self-paced objective function. While this implicit objective could provide helpful interpretations to the effectiveness, especially the robustness, insights under the SPL paradigms, there are still no theoretical results strictly proved to verify such relationship. To this issue, in this paper, we provide some convergence results on this implicit objective of SPL. Specifically, we prove that the learning process of SPL always converges to critical points of this implicit objective under some mild conditions. This result verifies the intrinsic relationship between SPL and this implicit objective, and makes the previous robustness analysis on SPL complete and theoretically rational.
@articleliu2018understanding, title=On convergence properties of implicit self-paced objective, author=Ma, Zilu and Liu, Shiqi and Meng, Deyu and Zhang, Yong and Lo, SioLong and Han, Zhi, journal=Information Sciences, year=2018, publisher=Elsevier
Auto robust relative radiometric normalization via latent change noise modelling

Shiqi Liu, Lu Wang, Jie Lian, Cong Liu, Xuchen Zhan, Jintao Lu, Jie Liu, Ting Wang, Dong Geng, Hongwei Duan, and others

arXiv preprint arXiv:2111.12406 , 2021

PDF Abs Bib

Relative radiometric normalization(RRN) of different satellite images of the same terrain is necessary for change detection, object classification/segmentation, and map-making tasks. However, traditional RRN models are not robust, disturbing by object change, and RRN models precisely considering object change can not robustly obtain the no-change set. This paper proposes auto robust relative radiometric normalization methods via latent change noise modeling. They utilize the prior knowledge that no change points possess small-scale noise under relative radiometric normalization and that change points possess large-scale radiometric noise after radiometric normalization, combining the stochastic expectation maximization method to quickly and robustly extract the no-change set to learn the relative radiometric normalization mapping functions. This makes our model theoretically grounded regarding the probabilistic theory and mathematics deduction. Specifically, when we select histogram matching as the relative radiometric normalization learning scheme integrating with the mixture of Gaussian noise(HM-RRN-MoG), the HM-RRN-MoG model achieves the best performance. Our model possesses the ability to robustly against clouds/fogs/changes. Our method naturally generates a robust evaluation indicator for RRN that is the no-change set root mean square error. We apply the HM-RRN-MoG model to the latter vegetation/water change detection task, which reduces the radiometric contrast and NDVI/NDWI differences on the no-change set, generates consistent and comparable results. We utilize the no-change set into the building change detection task, efficiently reducing the pseudo-change and boosting the precision.
@articleliu2021auto, title=Auto robust relative radiometric normalization via latent change noise modelling, author=Liu, Shiqi and Wang, Lu and Lian, Jie and Liu, Cong and Zhan, Xuchen and Lu, Jintao and Liu, Jie and Wang, Ting and Geng, Dong and Duan, Hongwei and others, journal=arXiv preprint arXiv:2111.12406, year=2021
Automatically eliminating seam lines with Poisson editing in complex relative radiometric normalization mosaicking scenarios

Shiqi Liu, Jie Lian, Xuchen Zhan, Cong Liu, Yuze Tian, and Hongwei Duan

arXiv preprint arXiv:2106.07441 , 2021

PDF Abs Bib

Relative radiometric normalization (RRN) mosaicking among multiple remote sensing images is crucial for the downstream tasks, including map-making, image recognition, semantic segmentation, and change detection. However, there are often seam lines on the mosaic boundary and radiometric contrast left, especially in complex scenarios, making the appearance of mosaic images unsightly and reducing the accuracy of the latter classification/recognition algorithms. This paper renders a novel automatical approach to eliminate seam lines in complex RRN mosaicking scenarios. It utilizes the histogram matching on the overlap area to alleviate radiometric contrast, Poisson editing to remove the seam lines, and merging procedure to determine the normalization transfer order. Our method can handle the mosaicking seam lines with arbitrary shapes and images with extreme topological relationships (with a small intersection area). These conditions make the main feathering or blending methods, e.g., linear weighted blending and Laplacian pyramid blending, unavailable. In the experiment, our approach visually surpasses the automatic methods without Poisson editing and the manual blurring and feathering method using GIMP software.
@articleliu2021automatically, title=Automatically eliminating seam lines with Poisson editing in complex relative radiometric normalization mosaicking scenarios, author=Liu, Shiqi and Lian, Jie and Zhan, Xuchen and Liu, Cong and Tian, Yuze and Duan, Hongwei, journal=arXiv preprint arXiv:2106.07441, year=2021
Understanding Pan-Sharpening via Generalized Inverse

Shiqi Liu, Yutong Bai, Xinyang Han, and Yuille Alan

arXiv preprint arXiv:2310.02718 , 2023

PDF Abs Bib

Pan-sharpening algorithm utilizes panchromatic image and multispectral image to obtain a high spatial and high spectral image. However, the optimizations of the algorithms are designed with different standards. We adopt the simple matrix equation to describe the Pan-sharpening problem. The solution existence condition and the acquirement of spectral and spatial resolution are discussed. A down-sampling enhancement method was introduced for better acquiring the spatial and spectral down-sample matrices. By the generalized inverse theory, we derived two forms of general inverse matrix formulations that can correspond to the two prominent classes of Pan-sharpening methods, that is, component substitution and multi-resolution analysis methods. Specifically, the Gram Schmidt Adaptive(GSA) was proved to follow the general inverse matrix formulation of component substitution. A model prior to the general inverse matrix of the spectral function was rendered. The theoretical errors are analyzed. Synthetic experiments and real data experiments are implemented. The proposed methods are better and sharper than other methods qualitatively in both synthetic and real experiments. The down-sample enhancement effect is shown of better results both quantitatively and qualitatively in real experiments. The generalized inverse matrix theory help us better understand the Pan-sharpening.
@articleliu2023understanding, title=Understanding Pan-Sharpening via Generalized Inverse, author=Liu, Shiqi and Bai, Yutong and Han, Xinyang and Yuille Alan, journal=arXiv preprint arXiv:2310.02718, year=2023
Geography Teaching
水蒸汽对气候变暖影响的实验报告

地理教学 (Geography Teaching) (Geography Teaching) , 2015

PDF Abs Bib

通过四次定性对比实验,一次模拟光源定量实验,对水蒸汽的吸热作用,温室效应进行了研究,用Excel软件将实验数据制作成实时温度变化曲线图,直观显示实验的结果.实验证明:在相同辐射条件下的密闭容器中,温度的上升与水蒸汽含量正相关;升温过程中,水蒸汽越多,越晚到达平衡温度,且平衡温度越高;不论在升温还是降温过程中,水蒸汽越多温度越高.
@articleliu2015Vapor, title=水蒸汽对气候变暖影响的实验报告, author=刘仕琪 (Shiqi Liu), 金广禹, 潘劭平, 马腾, 刘宇曼, 卢涵, 陈佑宁, 陈式如, journal=地理教学 (Geography Teaching), year=2015, publisher=