Yang Liu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Yang Liu has a total of 235 co-authors. Here are the top co-authors by number of shared publications:
Zhaowei Zhu
8 shared publicationsJialu Wang
✩ University of California Santa Cruz
6 shared publications- Mingyan Liu
University of Michigan
4 shared publications - Parinaz Naghizadeh
University of California, San Diego, USA
4 shared publications - Reilly Raab
Computer Science and Engineering University of California Santa Cruz 95064 Santa Cruz CA
4 shared publications - Yiling Chen
Harvard University
4 shared publications
Latest Publications
- 1 May 2026
- Published by Springer Nature
- p. 1-18
An Introduction to Machine Unlearning
- Open Access
Small-Margin Preferences Still Matter-If You Train Them Right
Preference optimization methods such as DPO align large language models (LLMs) using paired comparisons, but their effectiveness can be highly sensitive to the quality and difficulty of preference pairs. A common heuristic treats small-margin (ambiguous) pairs as noisy and filters them out. In this paper, we revisit this assumption and show that pair difficulty interacts strongly with the optimization objective: when trained with preference-based losses, difficult pairs can destabilize training and harm alignment, yet these same pairs still contain useful supervision signals when optimized with supervised fine-tuning (SFT). Motivated by this observation, we propose MixDPO, a simple yet effective difficulty-aware training strategy that (i) orders preference data from easy to hard (a curriculum over margin-defined difficulty), and (ii) routes difficult pairs to an SFT objective while applying a preference loss to easy pairs. This hybrid design provides a practical mechanism to leverage ambiguous pairs without incurring the optimization failures often associated with preference losses on low-margin data. Across three LLM-judge benchmarks, MixDPO consistently improves alignment over DPO and a range of widely-used variants, with particularly strong gains on AlpacaEval~2 length-controlled (LC) win rate.
Open AccessSmall-Margin Preferences Still Matter-If You Train Them Right
Preference optimization methods such as DPO align large language models (LLMs) using paired comparisons, but their effectiveness can be highly sensitive to the quality and difficulty of preference pairs. A common heuristic treats small-margin (ambiguous) pairs as noisy and filters them out. In this paper, we revisit this assumption and show that pair difficulty interacts strongly with the optimization objective: when trained with preference-based losses, difficult pairs can destabilize training and harm alignment, yet these same pairs still contain useful supervision signals when optimized with supervised fine-tuning (SFT). Motivated by this observation, we propose MixDPO, a simple yet effective difficulty-aware training strategy that (i) orders preference data from easy to hard (a curriculum over margin-defined difficulty), and (ii) routes difficult pairs to an SFT objective while applying a preference loss to easy pairs. This hybrid design provides a practical mechanism to leverage ambiguous pairs without incurring the optimization failures often associated with preference losses on low-margin data. Across three LLM-judge benchmarks, MixDPO consistently improves alignment over DPO and a range of widely-used variants, with particularly strong gains on AlpacaEval~2 length-controlled (LC) win rate.
- Open Access
Stabilizing Self-Consuming Diffusion Models with Latent Space Filtering
As synthetic data proliferates across the Internet, it is often reused to train successive generations of generative models. This creates a ``self-consuming loop" that can lead to training instability or \textit{model collapse}. Common strategies to address the issue -- such as accumulating historical training data or injecting fresh real data -- either increase computational cost or require expensive human annotation. In this paper, we empirically analyze the latent space dynamics of self-consuming diffusion models and observe that the low-dimensional structure of latent representations extracted from synthetic data degrade over generations. Based on this insight, we propose \textit{Latent Space Filtering} (LSF), a novel approach that mitigates model collapse by filtering out less realistic synthetic data from mixed datasets. Theoretically, we present a framework that connects latent space degradation to empirical observations. Experimentally, we show that LSF consistently outperforms existing baselines across multiple real-world datasets, effectively mitigating model collapse without increasing training cost or relying on human annotation.
Open AccessStabilizing Self-Consuming Diffusion Models with Latent Space Filtering
As synthetic data proliferates across the Internet, it is often reused to train successive generations of generative models. This creates a ``self-consuming loop" that can lead to training instability or \textit{model collapse}. Common strategies to address the issue -- such as accumulating historical training data or injecting fresh real data -- either increase computational cost or require expensive human annotation. In this paper, we empirically analyze the latent space dynamics of self-consuming diffusion models and observe that the low-dimensional structure of latent representations extracted from synthetic data degrade over generations. Based on this insight, we propose \textit{Latent Space Filtering} (LSF), a novel approach that mitigates model collapse by filtering out less realistic synthetic data from mixed datasets. Theoretically, we present a framework that connects latent space degradation to empirical observations. Experimentally, we show that LSF consistently outperforms existing baselines across multiple real-world datasets, effectively mitigating model collapse without increasing training cost or relying on human annotation.
- Open Access
- 20 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 1551-1562
Noise-Resilient Point-wise Anomaly Detection in Time Series Using Weak Segment Labels
- 20 July 2025
- Published by Association for Computing Machinery (ACM)
- p. 1551-1562
Open AccessNoise-Resilient Point-wise Anomaly Detection in Time Series Using Weak Segment Labels
- Open Access
Incentivizing High-quality Participation From Federated Learning Agents
Federated learning (FL) provides a promising paradigm for facilitating collaboration between multiple clients that jointly learn a global model without directly sharing their local data. However, existing research suffers from two caveats: 1) From the perspective of agents, voluntary and unselfish participation is often assumed. But self-interested agents may opt out of the system or provide low-quality contributions without proper incentives; 2) From the mechanism designer's perspective, the aggregated models can be unsatisfactory as the existing game-theoretical federated learning approach for data collection ignores the potential heterogeneous effort caused by contributed data. To alleviate above challenges, we propose an incentive-aware framework for agent participation that considers data heterogeneity to accelerate the convergence process. Specifically, we first introduce the notion of Wasserstein distance to explicitly illustrate the heterogeneous effort and reformulate the existing upper bound of convergence. To induce truthful reporting from agents, we analyze and measure the generalization error gap of any two agents by leveraging the peer prediction mechanism to develop score functions. We further present a two-stage Stackelberg game model that formalizes the process and examines the existence of equilibrium. Extensive experiments on real-world datasets demonstrate the effectiveness of our proposed mechanism.
Open AccessIncentivizing High-quality Participation From Federated Learning Agents
Federated learning (FL) provides a promising paradigm for facilitating collaboration between multiple clients that jointly learn a global model without directly sharing their local data. However, existing research suffers from two caveats: 1) From the perspective of agents, voluntary and unselfish participation is often assumed. But self-interested agents may opt out of the system or provide low-quality contributions without proper incentives; 2) From the mechanism designer's perspective, the aggregated models can be unsatisfactory as the existing game-theoretical federated learning approach for data collection ignores the potential heterogeneous effort caused by contributed data. To alleviate above challenges, we propose an incentive-aware framework for agent participation that considers data heterogeneity to accelerate the convergence process. Specifically, we first introduce the notion of Wasserstein distance to explicitly illustrate the heterogeneous effort and reformulate the existing upper bound of convergence. To induce truthful reporting from agents, we analyze and measure the generalization error gap of any two agents by leveraging the peer prediction mechanism to develop score functions. We further present a two-stage Stackelberg game model that formalizes the process and examines the existence of equilibrium. Extensive experiments on real-world datasets demonstrate the effectiveness of our proposed mechanism.
- Open Access
COSMIC: Generalized Refusal Direction Identification in LLM Activations
Large Language Models (LLMs) encode behaviors such as refusal within their activation space, yet identifying these behaviors remains a significant challenge. Existing methods often rely on predefined refusal templates detectable in output tokens or require manual analysis. We introduce \textbf{COSMIC} (Cosine Similarity Metrics for Inversion of Concepts), an automated framework for direction selection that identifies viable steering directions and target layers using cosine similarity - entirely independent of model outputs. COSMIC achieves steering performance comparable to prior methods without requiring assumptions about a model's refusal behavior, such as the presence of specific refusal tokens. It reliably identifies refusal directions in adversarial settings and weakly aligned models, and is capable of steering such models toward safer behavior with minimal increase in false refusals, demonstrating robustness across a wide range of alignment conditions.
Open AccessCOSMIC: Generalized Refusal Direction Identification in LLM Activations
Large Language Models (LLMs) encode behaviors such as refusal within their activation space, yet identifying these behaviors remains a significant challenge. Existing methods often rely on predefined refusal templates detectable in output tokens or require manual analysis. We introduce \textbf{COSMIC} (Cosine Similarity Metrics for Inversion of Concepts), an automated framework for direction selection that identifies viable steering directions and target layers using cosine similarity - entirely independent of model outputs. COSMIC achieves steering performance comparable to prior methods without requiring assumptions about a model's refusal behavior, such as the presence of specific refusal tokens. It reliably identifies refusal directions in adversarial settings and weakly aligned models, and is capable of steering such models toward safer behavior with minimal increase in false refusals, demonstrating robustness across a wide range of alignment conditions.
- 21 April 2025
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in INFORMS Journal on Computing
Adaptive Bounded Exploration and Intermediate Actions for Data Debiasing
The performance of algorithmic decision rules is largely dependent on the quality of the training data sets available to them. Biases in these data sets can raise economic and ethical concerns because of the disparate treatment of different groups by the resulting algorithms. In this paper, we propose algorithms for sequentially debiasing the training data set through adaptive and bounded exploration in a classification problem with costly and censored feedback. Our proposed algorithms balance between the ultimate goal of mitigating the impacts of data biases, which will, in turn, lead to more accurate and fairer decisions, and the exploration risks incurred to achieve this goal. Specifically, we propose adaptive bounds to limit the region of exploration and leverage intermediate actions, which provide noisy label information at a lower cost. We analytically show that such exploration can help debias data in certain distributions, investigate how algorithmic fairness interventions can work in conjunction with our proposed algorithms, and validate the performance of these algorithms through numerical experiments on synthetic and real-world data. History: Accepted by Kaushik Dutta, Area Editor for Responsible AI and Data Science for Social Good, for the INFORMS Journal on Computing Special Issue on Responsible AI and Data Science for Social Goods. Funding: This work is supported by the National Science Foundation (NSF) program on Fairness in AI in collaboration with Amazon under Award No. IIS-2040800, and by Cisco Research. Supplemental Material: The software that supports the findings of this study is available within the paper and its Supplemental Information ( https://pubsonline.informs.org/doi/suppl/10.1287/ijoc.2024.0651 ) as well as from the IJOC GitHub software repository ( https://github.com/INFORMSJoC/2024.0651 ). The complete IJOC Software and Data Repository is available at https://informsjoc.github.io/ .
- 21 April 2025
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in INFORMS Journal on Computing
Adaptive Bounded Exploration and Intermediate Actions for Data Debiasing
The performance of algorithmic decision rules is largely dependent on the quality of the training data sets available to them. Biases in these data sets can raise economic and ethical concerns because of the disparate treatment of different groups by the resulting algorithms. In this paper, we propose algorithms for sequentially debiasing the training data set through adaptive and bounded exploration in a classification problem with costly and censored feedback. Our proposed algorithms balance between the ultimate goal of mitigating the impacts of data biases, which will, in turn, lead to more accurate and fairer decisions, and the exploration risks incurred to achieve this goal. Specifically, we propose adaptive bounds to limit the region of exploration and leverage intermediate actions, which provide noisy label information at a lower cost. We analytically show that such exploration can help debias data in certain distributions, investigate how algorithmic fairness interventions can work in conjunction with our proposed algorithms, and validate the performance of these algorithms through numerical experiments on synthetic and real-world data. History: Accepted by Kaushik Dutta, Area Editor for Responsible AI and Data Science for Social Good, for the INFORMS Journal on Computing Special Issue on Responsible AI and Data Science for Social Goods. Funding: This work is supported by the National Science Foundation (NSF) program on Fairness in AI in collaboration with Amazon under Award No. IIS-2040800, and by Cisco Research. Supplemental Material: The software that supports the findings of this study is available within the paper and its Supplemental Information ( https://pubsonline.informs.org/doi/suppl/10.1287/ijoc.2024.0651 ) as well as from the IJOC GitHub software repository ( https://github.com/INFORMSJoC/2024.0651 ). The complete IJOC Software and Data Repository is available at https://informsjoc.github.io/ .
- Open Access
Adaptive Bounded Exploration and Intermediate Actions for Data Debiasing
The performance of algorithmic decision rules is largely dependent on the quality of training datasets available to them. Biases in these datasets can raise economic and ethical concerns due to the resulting algorithms' disparate treatment of different groups. In this paper, we propose algorithms for sequentially debiasing the training dataset through adaptive and bounded exploration in a classification problem with costly and censored feedback. Our proposed algorithms balance between the ultimate goal of mitigating the impacts of data biases -- which will in turn lead to more accurate and fairer decisions, and the exploration risks incurred to achieve this goal. Specifically, we propose adaptive bounds to limit the region of exploration, and leverage intermediate actions which provide noisy label information at a lower cost. We analytically show that such exploration can help debias data in certain distributions, investigate how {algorithmic fairness interventions} can work in conjunction with our proposed algorithms, and validate the performance of these algorithms through numerical experiments on synthetic and real-world data.
Open AccessAdaptive Bounded Exploration and Intermediate Actions for Data Debiasing
The performance of algorithmic decision rules is largely dependent on the quality of training datasets available to them. Biases in these datasets can raise economic and ethical concerns due to the resulting algorithms' disparate treatment of different groups. In this paper, we propose algorithms for sequentially debiasing the training dataset through adaptive and bounded exploration in a classification problem with costly and censored feedback. Our proposed algorithms balance between the ultimate goal of mitigating the impacts of data biases -- which will in turn lead to more accurate and fairer decisions, and the exploration risks incurred to achieve this goal. Specifically, we propose adaptive bounds to limit the region of exploration, and leverage intermediate actions which provide noisy label information at a lower cost. We analytically show that such exploration can help debias data in certain distributions, investigate how {algorithmic fairness interventions} can work in conjunction with our proposed algorithms, and validate the performance of these algorithms through numerical experiments on synthetic and real-world data.
- Open Access
- 11 April 2025
- Published by Association for the Advancement of Artificial Intelligence (AAAI) in Proceedings of the AAAI Conference on Artificial Intelligence
- Vol. 39 (13) , 14318-14326
Human and AI Perceptual Differences in Image Classification Errors
Artificial intelligence (AI) models for computer vision trained with supervised machine learning are assumed to solve classification tasks by imitating human behavior learned from training labels. Most efforts in recent vision research focus on measuring the model task performance using standardized benchmarks such as accuracy. However limited work has sought to understand the perceptual difference between humans and machines. To fill this gap, this study first analyzes the statistical distributions of mistakes from the two sources, and then explores how task difficulty level affects these distributions. We find that even when AI learns an excellent model from the training data, one that outperforms humans in overall accuracy, these AI models have significant and consistent differences from human perception. We demonstrate the importance of studying these differences with a simple human-AI teaming algorithm that outperforms humans alone, AI alone, or AI-AI teaming.
- 11 April 2025
- Published by Association for the Advancement of Artificial Intelligence (AAAI) in Proceedings of the AAAI Conference on Artificial Intelligence
- Vol. 39 (13) , 14318-14326
Open AccessHuman and AI Perceptual Differences in Image Classification Errors
Artificial intelligence (AI) models for computer vision trained with supervised machine learning are assumed to solve classification tasks by imitating human behavior learned from training labels. Most efforts in recent vision research focus on measuring the model task performance using standardized benchmarks such as accuracy. However limited work has sought to understand the perceptual difference between humans and machines. To fill this gap, this study first analyzes the statistical distributions of mistakes from the two sources, and then explores how task difficulty level affects these distributions. We find that even when AI learns an excellent model from the training data, one that outperforms humans in overall accuracy, these AI models have significant and consistent differences from human perception. We demonstrate the importance of studying these differences with a simple human-AI teaming algorithm that outperforms humans alone, AI alone, or AI-AI teaming.
- 27 March 2025
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in INFORMS Journal on Computing
Code and Data Repository for Adaptive Bounded Exploration and Intermediate Actions for Data Debiasing
The data in this repository is a snapshot of the data that was used in the research reported on in the paper Adaptive Bounded Exploration and Intermediate Actions for Data Debiasing by Yifan Yang, Yang Liu, and Parinaz Naghizadeh.
- 27 March 2025
- Published by Institute for Operations Research and the Management Sciences (INFORMS) in INFORMS Journal on Computing
Code and Data Repository for Adaptive Bounded Exploration and Intermediate Actions for Data Debiasing
The data in this repository is a snapshot of the data that was used in the research reported on in the paper Adaptive Bounded Exploration and Intermediate Actions for Data Debiasing by Yifan Yang, Yang Liu, and Parinaz Naghizadeh.