CS Student Seminar

Date	Speaker	Title
09/16	Shivanjali Ranashing	Role-based Community Detection in Directed Signed Graphs
		Abstract: Community detection in directed signed graphs is complex due to the added information encoded in edge direction and sign. This paper introduces a method that reconstructs a directed signed graph into a tripartite undirected graph, where each node plays three roles: source, positive target, and negative target. By embedding this tripartite graph using a random walks method, structural equivalence is preserved. Dense regions of source nodes are identified as seed communities, which are expanded by including closely connected nodes of other roles. This role-based approach results in meaningful community structures, validated through experiments on test datasets. Bio: Shivanjali is a phd student working with Dr. Raj Bhatnagar.
09/23	Po-Yu Liang	Exploring Latent Space for Generating Peptide Analogs Using Protein Language Models
		Abstract: Generating short chains of amino acids, known as peptides, with specific properties is crucial in drug discovery and biotechnology. However, traditional methods often rely on large datasets, limiting their effectiveness. In this work, we introduce a new approach that uses autoencoder-like models to explore protein embedding spaces and create new variations of peptides using protein language models. Unlike conventional methods, our approach only requires a single sequence, eliminating the need for large datasets. Results show improved performance in structure and bioactivity similarity. Validated with Molecular Dynamics simulations on TIGIT, a cancer-related protein, our method generates unique peptides, offering promising potential for peptide screening. Bio: Po-Yu is a phd student working with Dr. Jun Bai.
09/30	Lucas Jividen	Uncovering the Mechanism of Hepatotoxicity of PFAS Targeting L-FABP Using GCN and Computational Modeling
		Abstract: Per- and polyfluoroalkyl substances (PFAS) are environmental pollutants with known toxicity and bioaccumulation issues. Their widespread industrial use and resistance to degradation have led to global environmental contamination and significant health concerns. While a minority of PFAS have been extensively studied, the toxicity of many PFAS remains poorly understood due to limited toxicological data. This study advances the predictive modeling of PFAS toxicity by utilizing machine learning, particularly graph convolutional networks (GCNs). Our results provide a more accurate ability to estimate PFAS toxicity to provide guidance in chemical discovery of new compounds and the development of new safety regulations. Bio: Lucas is a phd student working with Dr. Jun Bai.
10/07	Bangyan Ju	A Study of Using Multimodal LLMs for Non-Crash Functional Bug Detection in Android Apps
		Abstract: Numerous approaches employing various strategies have been developed to test the graphical user interfaces (GUIs) of mobile apps. However, traditional GUI testing techniques, such as random and model-based testing, primarily focus on generating test sequences that excel in achieving high code coverage but often fail to act as effective test oracles for non-crash functional (NCF) bug detection. To tackle these limitations, this study empirically investigates the capability of leveraging large language models (LLMs) to be test oracles to detect NCF bugs in Android apps. The results demonstrated that LLMs achieve a 49% bug detection rate, outperforming existing tools for detecting NCF bugs in Android apps. Bio: Bangyan is a phd student working with Dr. Yu Zhao.
10/21	Usman Anjum	A Novel Tsetlin Machine with Enhanced Generalization
		Abstract: The Tsetlin Machine (TM) is a novel machine learning approach that implements propositional logic to perform various tasks such as classification and regression. The TM not only achieves competitive accuracy in these tasks but also provides results that are explainable and easy to implement using simple hardware. The TM learns using clauses based on the features of the data, and final classification is done using a combination of these clauses. In this paper, we propose the novel idea of adding regularizers to the TM, referred to as Regularized TM (RegTM), to improve generalization. Regularizers have been widely used in machine learning to enhance accuracy. We explore different regularization strategies and their influence on performance. We demonstrate the feasibility of our methodology through various experiments on benchmark datasets. Bio: Usman is a postdoc researcher working with Dr. Justin Zhan.
10/28	Saptarshi Ghosh	Metonymy Resolution with Large Language Models
		Abstract: Metonymy plays an important role in our daily communication. People naturally think about things using their most salient properties or commonly related concepts. For example, by saying ''The bus decided to skip our stop today'', we actually mean that the bus driver made the decision, not the bus. Prior work on metonymy resolution has mainly focused on named entities. However, metonymy involving common nouns is also a common and challenging phenomenon. We argue that NLP systems should be capable of identifying the metonymic use of common nouns in context. In this presentation, I will introduce how we defined the problem, created our own dataset, and built models using large language models to tackle the task. Bio: Saptarshi is a MS student working with Dr. Tianyu Jiang.
11/04	Sakshi Singh	Similarity Aware Semi Supervised Time Series Classification
		Abstract: Accurate classification of Time-series data is essential for applications in healthcare, manufacturing, and finance. Existing deep learning methods have a limitation -- they struggle when labeled data is limited. We will be presenting SimTSC, a novel approach, that offers a new solution by integrating two key approaches: measuring similarity between time-series and deep learning for prediction. It uses graphs to connect labeled and unlabeled data based on similarities, enhancing the model's ability to learn patterns with minimal labels as this approach enables the model to learn from both labeled and unlabeled data using graph neural networks (GNNs). This approach enables the model to better capture complex patterns, improving performance even with minimal labeled examples. Bio: Sakshi is a MS student working with Dr. Gowtham Atluri.
11/18	Iman Chatterjee	Recognition of Dyadic Conversation States using Physiological Responses and Machine Learning
		Abstract: Two interacting individuals mirror one another’s behavior in numerous ways like unconscious gesture imitation or similar choice of words for expression. Such convergent behavior even exhibits itself at a subconscious level: physiological responses of the two may become similar as they keep interacting. Analyzing these synchronous behavior patterns can yield vital information about emotions exchanged during the interaction. Affective computing is the area of study focusing on design and implementation of systems that can interpret, process and influence human emotion. To build such systems, we rely on physiological responses of the two interacting individuals. Since noise contamination hinders interpretation of raw physiological signals, machine learning (ML) techniques are used to derive more robust and interpretable measures from the signal data. Furthermore, we also combine the multiple signal modalities namely, ECG, respiration patterns, galvanic skin reactance and skin temperature for our analysis. Bio: Iman is a phd student working with Dr. Vesna D. Novak.