arXiv subjects

Siyuan Tang

Publications and source records attributed to Siyuan Tang.

18 recordsLinked to original sources

arXiv2026

Conditional-Independence-Regularized Distributional Autoencoders for Mixed-Type Data

Mixed-type data containing both numerical and categorical variables arise in many scientific and real-world applications. Existing representation learning and generative modeling approaches typically focus either on reconstruction accuracy or unconditional data generation, but often fail to recover the full conditional distribution of the data while preserving interpretable structural relationships between heterogeneous variable types. In this work, we introduce Conditional-Independence-Regularized Distributional Autoencoders, a framework for learning low-dimensional representations of mixed-type data through conditional distribution matching and structural regularization. Our method combines an energy-score-based objective for numerical variables, a likelihood-based objective for categorical variables, and an auxiliary conditional independence regularization term encouraging the learned representation to capture the dependence between numerical and categorical components. We provide theoretical analysis showing that the optimal representation balances unexplained numerical variability, conditional entropy of categorical variables, and residual conditional dependence. Empirically, the proposed method achieves strong performance on both synthetic and real-world datasets, substantially improving categorical distribution recovery, achieving competitive overall conditional distribution recovery, and preserving mixed-type dependence structure. The code has been made available at GitHub.

stat.ME↗

arXiv2026

PrivUn: Unveiling Latent Ripple Effects and Shallow Forgetting in Privacy Unlearning

Large language models (LLMs) often memorize private information during training, raising serious privacy concerns. While machine unlearning has emerged as a promising solution, its true effectiveness against privacy attacks remains unclear. To address this, we propose PrivUn, a new evaluation framework that systematically assesses unlearning robustness through three-tier attack scenarios: direct retrieval, in-context learning recovery, and fine-tuning restoration; combined with quantitative analysis using forgetting scores, association metrics, and forgetting depth assessment. Our study exposes significant weaknesses in current unlearning methods, revealing two key findings: 1) unlearning exhibits gradient-driven ripple effects: unlike traditional forgetting which follows semantic relations (e.g., knowledge graphs), privacy unlearning propagates across latent gradient-based associations; and 2) most methods suffer from shallow forgetting, failing to remove private information distributed across multiple deep model layers. To validate these insights, we explore two strategies: association-aware core-set selection that leverages gradient similarity, and multi-layer deep intervention through representational constraints. These strategies represent a paradigm shift from shallow forgetting to deep forgetting.

cs.LG↗

arXiv2025

Natural Language-based Assessment of L2 Oral Proficiency using LLMs

Natural language-based assessment (NLA) is an approach to second language assessment that uses instructions - expressed in the form of can-do descriptors - originally intended for human examiners, aiming to determine whether large language models (LLMs) can interpret and apply them in ways comparable to human assessment. In this work, we explore the use of such descriptors with an open-source LLM, Qwen 2.5 72B, to assess responses from the publicly available S&I Corpus in a zero-shot setting. Our results show that this approach - relying solely on textual information - achieves competitive performance: while it does not outperform state-of-the-art speech LLMs fine-tuned for the task, it surpasses a BERT-based model trained specifically for this purpose. NLA proves particularly effective in mismatched task settings, is generalisable to other data types and languages, and offers greater interpretability, as it is grounded in clearly explainable, widely applicable language descriptors.

eess.AS↗

arXiv2025

Assessment of L2 Oral Proficiency using Speech Large Language Models

The growing population of L2 English speakers has increased the demand for developing automatic graders for spoken language assessment (SLA). Historically, statistical models, text encoders, and self-supervised speech models have been utilised for this task. However, cascaded systems suffer from the loss of information, while E2E graders also have limitations. With the recent advancements of multi-modal large language models (LLMs), we aim to explore their potential as L2 oral proficiency graders and overcome these issues. In this work, we compare various training strategies using regression and classification targets. Our results show that speech LLMs outperform all previous competitive baselines, achieving superior performance on two datasets. Furthermore, the trained grader demonstrates strong generalisation capabilities in the cross-part or cross-task evaluation, facilitated by the audio understanding knowledge acquired during LLM pre-training.

cs.CL↗

arXiv2025

Effective Exponential Drifts on Strata of Abelian Differentials

We study the dynamics of $SL_{2}(\mathbb{R})$ on the stratum of translation surfaces $\mathcal{H}(2)$. In particular, we prove that an orbit of the upper triangular subgroup of $SL_{2}(\mathbb{R})$ has a discretized dimension of almost $1$ in a direction transverse to the $SL_{2}(\mathbb{R})$-orbit. The proof proceeds via an effective closing lemma, and the Margulis function technique, which serves as an effective version of the exponential drift on $\mathcal{H}(2)$. The idea is based on the use of McMullen's classification theorem, together with Lindenstrauss-Mohammadi-Wang's effective equidistribution theorems in homogeneous dynamics.

math.DS↗

arXiv2024

Speak & Improve Challenge 2025: Tasks and Baseline Systems

This paper presents the "Speak & Improve Challenge 2025: Spoken Language Assessment and Feedback" -- a challenge associated with the ISCA SLaTE 2025 Workshop. The goal of the challenge is to advance research on spoken language assessment and feedback, with tasks associated with both the underlying technology and language learning feedback. Linked with the challenge, the Speak & Improve (S&I) Corpus 2025 is being pre-released, a dataset of L2 learner English data with holistic scores and language error annotation, collected from open (spontaneous) speaking tests on the Speak & Improve learning platform. The corpus consists of approximately 315 hours of audio data from second language English learners with holistic scores, and a 55-hour subset with manual transcriptions and error labels. The Challenge has four shared tasks: Automatic Speech Recognition (ASR), Spoken Language Assessment (SLA), Spoken Grammatical Error Correction (SGEC), and Spoken Grammatical Error Correction Feedback (SGECF). Each of these tasks has a closed track where a predetermined set of models and data sources are allowed to be used, and an open track where any public resource may be used. Challenge participants may do one or more of the tasks. This paper describes the challenge, the S&I Corpus 2025, and the baseline systems released for the Challenge.

cs.CL↗

arXiv2024

LDACP: Long-Delayed Ad Conversions Prediction Model for Bidding Strategy

In online advertising, once an ad campaign is deployed, the automated bidding system dynamically adjusts the bidding strategy to optimize Cost Per Action (CPA) based on the number of ad conversions. For ads with a long conversion delay, relying solely on the real-time tracked conversion number as a signal for bidding strategy can significantly overestimate the current CPA, leading to conservative bidding strategies. Therefore, it is crucial to predict the number of long-delayed conversions. Nonetheless, it is challenging to predict ad conversion numbers through traditional regression methods due to the wide range of ad conversion numbers. Previous regression works have addressed this challenge by transforming regression problems into bucket classification problems, achieving success in various scenarios. However, specific challenges arise when predicting the number of ad conversions: 1) The integer nature of ad conversion numbers exacerbates the discontinuity issue in one-hot hard labels; 2) The long-tail distribution of ad conversion numbers complicates tail data prediction. In this paper, we propose the Long-Delayed Ad Conversions Prediction model for bidding strategy (LDACP), which consists of two sub-modules. To alleviate the issue of discontinuity in one-hot hard labels, the Bucket Classification Module with label Smoothing method (BCMS) converts one-hot hard labels into non-normalized soft labels, then fits these soft labels by minimizing classification loss and regression loss. To address the challenge of predicting tail data, the Value Regression Module with Proxy labels (VRMP) uses the prediction bias of aggregated pCTCVR as proxy labels. Finally, a Mixture of Experts (MoE) structure integrates the predictions from BCMS and VRMP to obtain the final predicted ad conversion number.

cs.LG↗

arXiv2024

Effective density of surfaces near Teichm\"{u}ller curves

We study the dynamics of $SL_{2}(\mathbb{R})$ on the stratum of translation surfaces $\mathcal{H}(2)$. Especially, we obtain effective density theorems on $\mathcal{H}(2)$ for orbits of the upper triangular subgroup $P$ of $SL_{2}(\mathbb{R})$ with the based surfaces near a small Teichm\"{u}ller curve. The proof is based on the use of McMullen's classification theorem, together with the effective equidistribution theorems in homogeneous dynamics. In particular, we compare the $P$-orbit of a surface, and the $P$-orbit of its absolute periods using the Lindenstrauss-Mohammadi-Wang's effective equidistribution theorem.

math.DS↗

arXiv2024

Breast tumor classification based on self-supervised contrastive learning from ultrasound videos

Background: Breast ultrasound is prominently used in diagnosing breast tumors. At present, many automatic systems based on deep learning have been developed to help radiologists in diagnosis. However, training such systems remains challenging because they are usually data-hungry and demand amounts of labeled data, which need professional knowledge and are expensive. Methods: We adopted a triplet network and a self-supervised contrastive learning technique to learn representations from unlabeled breast ultrasound video clips. We further designed a new hard triplet loss to to learn representations that particularly discriminate positive and negative image pairs that are hard to recognize. We also constructed a pretraining dataset from breast ultrasound videos (1,360 videos from 200 patients), which includes an anchor sample dataset with 11,805 images, a positive sample dataset with 188,880 images, and a negative sample dataset dynamically generated from video clips. Further, we constructed a finetuning dataset, including 400 images from 66 patients. We transferred the pretrained network to a downstream benign/malignant classification task and compared the performance with other state-of-the-art models, including three models pretrained on ImageNet and a previous contrastive learning model retrained on our datasets. Results and conclusion: Experiments revealed that our model achieved an area under the receiver operating characteristic curve (AUC) of 0.952, which is significantly higher than the others. Further, we assessed the dependence of our pretrained model on the number of labeled data and revealed that <100 samples were required to achieve an AUC of 0.901. The proposed framework greatly reduces the demand for labeled data and holds potential for use in automatic breast ultrasound image diagnosis.

cs.CV↗

arXiv2024

Learn and Don't Forget: Adding a New Language to ASR Foundation Models

Foundation ASR models often support many languages, e.g. 100 languages in Whisper. However, there has been limited work on integrating an additional, typically low-resource, language, while maintaining performance on the original language set. Fine-tuning, while simple, may degrade the accuracy of the original set. We compare three approaches that exploit adaptation parameters: soft language code tuning, train only the language code; soft prompt tuning, train prepended tokens; and LoRA where a small set of additional parameters are optimised. Elastic Weight Consolidation (EWC) offers an alternative compromise with the potential to maintain performance in specific target languages. Results show that direct fine-tuning yields the best performance for the new language but degrades existing language capabilities. EWC can address this issue for specific languages. If only adaptation parameters are used, the language capabilities are maintained but at the cost of performance in the new language.

eess.AS↗

arXiv2024

Cross-Lingual Transfer Learning for Speech Translation

There has been increasing interest in building multilingual foundation models for NLP and speech research. This paper examines how to expand the speech translation capability of these models with restricted data. Whisper, a speech foundation model with strong performance on speech recognition and English translation, is used as the example model. Using speech-to-speech retrieval to analyse the audio representations generated by the encoder, we show that utterances from different languages are mapped to a shared semantic space. This shared embedding space can then be leveraged for zero-shot cross-lingual transfer in speech translation. By fine-tuning the Whisper decoder with only English-to-Chinese speech translation data, improved performance for translation to Chinese can be obtained for multiple languages, in addition to English. Furthermore, for languages related to those seen in training it is possible to perform speech translation, despite the model never seeing the language in training, or being able to perform transcription.

cs.CL↗

arXiv2023

The Janus Interface: How Fine-Tuning in Large Language Models Amplifies the Privacy Risks

The rapid advancements of large language models (LLMs) have raised public concerns about the privacy leakage of personally identifiable information (PII) within their extensive training datasets. Recent studies have demonstrated that an adversary could extract highly sensitive privacy data from the training data of LLMs with carefully designed prompts. However, these attacks suffer from the model's tendency to hallucinate and catastrophic forgetting (CF) in the pre-training stage, rendering the veracity of divulged PIIs negligible. In our research, we propose a novel attack, Janus, which exploits the fine-tuning interface to recover forgotten PIIs from the pre-training data in LLMs. We formalize the privacy leakage problem in LLMs and explain why forgotten PIIs can be recovered through empirical analysis on open-source language models. Based upon these insights, we evaluate the performance of Janus on both open-source language models and two latest LLMs, i.e., GPT-3.5-Turbo and LLaMA-2-7b. Our experiment results show that Janus amplifies the privacy risks by over 10 times in comparison with the baseline and significantly outperforms the state-of-the-art privacy extraction attacks including prefix attacks and in-context learning (ICL). Furthermore, our analysis validates that existing fine-tuning APIs provided by OpenAI and Azure AI Studio are susceptible to our Janus attack, allowing an adversary to conduct such an attack at a low cost.

cs.CR↗

arXiv2023

Gradient Shaping: Enhancing Backdoor Attack Against Reverse Engineering

Most existing methods to detect backdoored machine learning (ML) models take one of the two approaches: trigger inversion (aka. reverse engineer) and weight analysis (aka. model diagnosis). In particular, the gradient-based trigger inversion is considered to be among the most effective backdoor detection techniques, as evidenced by the TrojAI competition, Trojan Detection Challenge and backdoorBench. However, little has been done to understand why this technique works so well and, more importantly, whether it raises the bar to the backdoor attack. In this paper, we report the first attempt to answer this question by analyzing the change rate of the backdoored model around its trigger-carrying inputs. Our study shows that existing attacks tend to inject the backdoor characterized by a low change rate around trigger-carrying inputs, which are easy to capture by gradient-based trigger inversion. In the meantime, we found that the low change rate is not necessary for a backdoor attack to succeed: we design a new attack enhancement called \textit{Gradient Shaping} (GRASP), which follows the opposite direction of adversarial training to reduce the change rate of a backdoored model with regard to the trigger, without undermining its backdoor effect. Also, we provide a theoretic analysis to explain the effectiveness of this new technique and the fundamental weakness of gradient-based trigger inversion. Finally, we perform both theoretical and experimental analysis, showing that the GRASP enhancement does not reduce the effectiveness of the stealthy attacks against the backdoor detection methods based on weight analysis, as well as other backdoor mitigation methods without using detection.

cs.CR↗

arXiv2022

Selective Amnesia: On Efficient, High-Fidelity and Blind Suppression of Backdoor Effects in Trojaned Machine Learning Models

In this paper, we present a simple yet surprisingly effective technique to induce "selective amnesia" on a backdoored model. Our approach, called SEAM, has been inspired by the problem of catastrophic forgetting (CF), a long standing issue in continual learning. Our idea is to retrain a given DNN model on randomly labeled clean data, to induce a CF on the model, leading to a sudden forget on both primary and backdoor tasks; then we recover the primary task by retraining the randomized model on correctly labeled clean data. We analyzed SEAM by modeling the unlearning process as continual learning and further approximating a DNN using Neural Tangent Kernel for measuring CF. Our analysis shows that our random-labeling approach actually maximizes the CF on an unknown backdoor in the absence of triggered inputs, and also preserves some feature extraction in the network to enable a fast revival of the primary task. We further evaluated SEAM on both image processing and Natural Language Processing tasks, under both data contamination and training manipulation attacks, over thousands of models either trained on popular image datasets or provided by the TrojAI competition. Our experiments show that SEAM vastly outperforms the state-of-the-art unlearning techniques, achieving a high Fidelity (measuring the gap between the accuracy of the primary task and that of the backdoor) within a few minutes (about 30 times faster than training a model from scratch using the MNIST dataset), with only a small amount of clean data (0.1% of training data for TrojAI models).

cs.LG↗

arXiv2022

Stealthy Peers: Understanding Security Risks of WebRTC-Based Peer-Assisted Video Streaming

As an emerging service for in-browser content delivery, peer-assisted delivery network (PDN) is reported to offload up to 95\% of bandwidth consumption for video streaming, significantly reducing the cost incurred by traditional CDN services. With such benefits, PDN services significantly impact today's video streaming and content delivery model. However, their security implications have never been investigated. In this paper, we report the first effort to address this issue, which is made possible by a suite of methodologies, e.g., an automatic pipeline to discover PDN services and their customers, and a PDN analysis framework to test the potential security and privacy risks of these services. Our study has led to the discovery of 3 representative PDN providers, along with 134 websites and 38 mobile apps as their customers. Most of these PDN customers are prominent video streaming services with millions of monthly visits or app downloads (from Google Play). Also found in our study are another 9 top video/live streaming websites with each equipped with a proprietary PDN solution. Most importantly, our analysis on these PDN services has brought to light a series of security risks, which have never been reported before, including free riding of the public PDN services, video segment pollution, exposure of video viewers' IPs to other peers, and resource squatting. All such risks have been studied through controlled experiments and measurements, under the guidance of our institution's IRB. We have responsibly disclosed these security risks to relevant PDN providers, who have acknowledged our findings, and also discussed the avenues to mitigate these risks.

cs.CR↗

arXiv2022

Clues in Tweets: Twitter-Guided Discovery and Analysis of SMS Spam

With its critical role in business and service delivery through mobile devices, SMS (Short Message Service) has long been abused for spamming, which is still on the rise today possibly due to the emergence of A2P bulk messaging. The effort to control SMS spam has been hampered by the lack of up-to-date information about illicit activities. In our research, we proposed a novel solution to collect recent SMS spam data, at a large scale, from Twitter, where users voluntarily report the spam messages they receive. For this purpose, we designed and implemented SpamHunter, an automated pipeline to discover SMS spam reporting tweets and extract message content from the attached screenshots. Leveraging SpamHunter, we collected from Twitter a dataset of 21,918 SMS spam messages in 75 languages, spanning over four years. To our best knowledge, this is the largest SMS spam dataset ever made public. More importantly, SpamHunter enables us to continuously monitor emerging SMS spam messages, which facilitates the ongoing effort to mitigate SMS spamming. We also performed an in-depth measurement study that sheds light on the new trends in the spammer's strategies, infrastructure and spam campaigns. We also utilized our spam SMS data to evaluate the robustness of the spam countermeasures put in place by the SMS ecosystem, including anti-spam services, bulk SMS services, and text messaging apps. Our evaluation shows that such protection cannot effectively handle those spam samples: either introducing significant false positives or missing a large number of newly reported spam messages.

cs.CR↗

arXiv2021

New time-changes of unipotent flows on quotients of Lorentz groups

We study the cocompact lattices $Γ\subset SO(n,1)$ so that the Laplace-Beltrami operator $Δ$ on $SO(n)\backslash SO(n,1)/Γ$ has eigenvalues in $(0,1/4)$, and then show that there exist time-changes of unipotent flows on $SO(n,1)/Γ$ that are not measurably conjugate to the unperturbed ones. A main ingredient of the proof is a stronger version of the branching of the complementary series. Combining it with a refinement of the works of Ratner and Flaminio-Forni is adequate for our purpose.

math.DS↗

arXiv2021

Rigidity of joinings for time-changes of unipotent flows on quotients of Lorentz groups

Let $u_{X}^{t}$ be a unipotent flow on $X=SO(n,1)/Γ$, $u_{Y}^{t}$ be a unipotent flow on $Y=G/Γ^{\prime}$. Let $\tilde{u}_{X}^{t}$, $\tilde{u}_{Y}^{t}$ be time-changes of $u_{X}^{t}$, $u_{Y}^{t}$ respectively. We show the disjointness (in the sense of Furstenberg) between $u_{X}^{t}$ and $\tilde{u}_{Y}^{t}$ (or $\tilde{u}_{X}^{t}$ and $u_{Y}^{t}$) in certain situations. Our method refines the works of Ratner and extends a recent work of Dong, Kanigowski and Wei.

math.DS↗