Nam Nguyen

4-5pm, Dec 17 2020

https://stonybrook.zoom.us/j/94214254415?pwd=K1VoQml4cFdlVW51VW41dWtid2tJdz09



The molecular mechanisms and functions in complex biological systems
currently remain elusive. Recent high-throughput techniques, such as
next-generation sequencing, have generated a wide variety of
multiomics datasets that enable the identification of biological
functions and mechanisms via multiple facets. However, integrating
these large-scale multiomics data and discovering functional insights
are, nevertheless, challenging tasks. To address these challenges,
machine learning has been broadly applied to analyze multiomics. In
particular, multiview learning is more effective than previous
integrative methods for learning data's heterogeneity and revealing
cross-talk patterns. Although it has been applied to various contexts,
such as computer vision and speech recognition, multiview learning has
not yet been widely applied to biological data--specifically,
multiomics data. Therefore, we have developed a framework called
multiview empirical risk minimization (MV-ERM) for unifying multiview
learning methods (Nguyen, et al., PLoS Computational Biology, 2020).
MV-ERM enables potential applications to understand multiomics
including genomics, transcriptomics, and epigenomics, in an aim to
discover the functional and mechanistic interpretations across omics.
Based on MV-ERM, we have developed the following methods:
ManiNetCluster, Varmole and ECMarker.



(1) ManiNetCluster (Nguyen, et al., BMC Genomics, 2019) is a manifold
learning method which simultaneously aligns and clusters gene networks
(e.g., co-expression) to systematically reveal the links of genomic
function between different phenotypes. Specifically, ManiNetCluster
employs manifold alignment to uncover and match local and non-linear
structures among networks, and identifies cross-network functional
links. We demonstrated that ManiNetCluster better aligns the
orthologous genes from their developmental expression profiles across
model organisms than state-of-the-art methods. This indicates the
potential non-linear interactions of evolutionarily conserved genes
across species in development. Furthermore, we applied ManiNetCluster
to time series transcriptome data measured in the green alga
Chlamydomonas reinhardtii to discover the genomic functions linking
various metabolic processes between the light and dark periods of a
diurnally cycling culture;



(2) Varmole (Nguyen, et al., Bioinformatics, 2020) is an interpretable
deep learning method that simultaneously reveals genomic functions and
mechanisms while predicting phenotype from genotype. In particular,
Varmole embeds multi-omic networks into a deep neural network
architecture and prioritizes variants, genes and regulatory linkages
via biological drop-connect without needing prior feature selections.
With an application to schizophonia, we demonstrate that Varmole
provides an effective alternative for recent statistical methods that
associate functional omic data (e.g. gene expression) with genotype
and phenotype and that link variants to individual genes in population
studies such as genome-wide association study;



(3) ECMarker (Jin*, Nguyen*, et al., Bioinformatics, 2020) is an
interpretable and scalable machine learning model that predicts gene
expression biomarkers for disease phenotypes and simultaneously
reveals underlying regulatory mechanisms. Particularly, ECMarker is
built on the integration of semi- and discriminative- restricted
Boltzmann machines, a neural network model for classification allowing
lateral connections at the input gene layer. With application to the
gene expression data of non-small cell lung cancer (NSCLC) patients,
we found that ECMarker not only achieved a relatively high accuracy
for predicting cancer stages but also identified the biomarker genes
and gene networks implying the regulatory mechanisms in lung cancer
development.



Finally, we propose a novel multiview learning method, Malignomics, to
predict phenotypes from heterogeneous multi-omic features. Malignomics
will first align multi-omic features by deep manifold alignment onto a
common latent space, better predicting nonlinear relationships across
omics. This deep alignment aims to preserve both global consistency
and local smoothness across omics and reveal higher-order nonlinear
interactions (i.e., manifolds) among cross-omic features. Second, it
uses these manifold structures to regularize the classifiers for
predicting phenotypes. This manifold-regularization allows
highlighting cross-omic feature manifolds and prioritizing the
features and interactions for the phenotypes. The prioritized
multi-omic features will further reveal underlying phenotypic
functions and mechanisms and thus enhance the biological
interpretation of Malignomics. We will apply Malignomics to
multi-omics data in neuropsychiatric disorders, and prioritize gene
regulatory networks linking risk variants, regulatory elements, and
genes for the disorders. We will also compare Malignomics with the
state-of-the-arts, and investigate how the manifold regulation will
potentially improve understanding of multi-omics functions and
predicting diseases.
The overall purpose of this seminar is to bring together people with interests in Computer Vision theory and techniques and to examine current research issues. This course will be appropriate for people who already took a Computer Vision graduate course or already had research experience in Computer Vision. To enroll in this course, you must either: (1) be in the PhD program or (2) receive permission from the instructors.

Each seminar will consist of multiple short talks (around 10 minutes) by multiple people. Students can register for 1 credit for CSE 656. Registered students must attend and present a minimum of 2 or 3 talks. Everyone else is welcome to attend. Fill in https://forms.gle/pCVXovgfMfQwGqG38 to subscribe to our mailing list for further announcement.
Hieu Le presents Incorporating Physical Illumination Constraints into Deep Learning Shadow Detection and Removal (PhD Proposal)

Shadows provide useful cues to analyze the scene but also hamper many computer vision algorithms such as image segmentation, object detection or tracking. For those reasons, shadow detection and shadow removal have been well studied topics in computer vision. Early approaches for shadow detection and removal focus on physical illumination models of shadows. These methods can express, identify, and remove shadows in a physically plausible manner. However, these models are often hard to optimize and slow in inference due to reliance on hand-designed image features. On the other hand, recent deep-learning approaches have achieved breakthroughs in performances for both shadow detection and removal. They learn to extract useful features automatically through training while being extremely efficient in computation. However, these models are data-dependent, opaque and ignore the physical aspects of shadows.

We propose to incorporate physical illumination constraints into deep-learning frameworks. Thus the mapping learned by the deep-network closely follows the physics of shadows, enabling the network to systematically and realistically modify shadows in images. For shadow detection, we present a novel GAN framework in which the generator can generate realistic images with attenuated shadows that can be used to train a shadow detector. For shadow removal, we propose a method that uses deep-networks to estimate the unknown parameters for a shadow image formation model that removes shadows. The system outputs shadow-free images in high-quality with no image artifacts and achieves state-of-the-art shadow removal performance. Lastly, we propose a system trained without the need for any shadow-free images in which physical constraints play pivotal roles that enable training the networks.

For Zoom information, please email events@cs.stonybrook.edu.
Abstract: DeepSeek-R1-Zero has shown that reinforcement learning (RL) at scale can directly enhance the reasoning capabilities of LLMs without supervised fine-tuning. In this work, we critically examine R1-Zero-like training by analyzing its two core components: base models and RL. We investigate a wide range of base models, including DeepSeek-V3-Base, to understand how pretraining characteristics influence RL performance. Our analysis reveals that DeepSeek-V3-Base already exhibit ''Aha moment'', while Qwen2.5 base models demonstrate strong reasoning capabilities even without prompt templates, suggesting potential pretraining biases. Additionally, we identify an optimization bias in Group Relative Policy Optimization (GRPO), which artificially increases response length (especially for incorrect outputs) during training. To address this, we introduce Dr. GRPO, an unbiased optimization method that improves token efficiency while maintaining reasoning performance. Leveraging these insights, we present a minimalist R1-Zero recipe that achieves 43.3% accuracy on AIME 2024 with a 7B base model, establishing a new state-of-the-art.

Speaker: Md. Saqib Hasan

Location: CS2311

Abstract: Millions of individuals living in disadvantaged communities are burdened by poverty, illegal drug activities, health concerns, and the lack of reliable and affordable access to facilities (e.g., schools, hospitals, and transit stations). To address these societal problems efficiently with broad support, initiatives have called to engage agents (e.g., residents, community leaders, or stakeholders) and consider their preferences on community improvement decisions to make collective community decisions. In this talk, we will focus on our ongoing AI-empowered collective decision-making approaches to improve the accessibility of individuals to facilities by (a) locating facilities to provide essential services and (b) strengthening existing infrastructures via structural modifications (e.g., constructing new roads, bridges, multi-use paths, or shuttle services) subject to individuals' preferences on the locations of the facilities and which communities to improve access, respectively. In particular, we will discuss our (theoretical and algorithmic) studies on modeling these approaches under several settings (e.g., accounting for fairness and agent preferences) and designing fair, transparent, strategy proof, and (approximately) optimal mechanisms to elicit (true) individual preferences and determine collective community decisions in order to improve facility accessibility. Finally, we will discuss other ongoing and future collective decision-making efforts in urban planning and public health (i.e., our recent studies on substance use research) to improve communities.

Bio: Hau Chan is an assistant professor in the School of Computing at the University of Nebraska-Lincoln. He received his Ph.D. in Computer Science from Stony Brook University in 2015 and completed three years of Postdoctoral Fellowships, including at the Laboratory for Innovation Science at Harvard University in 2018. His main research lies in multi-agent aspects of AI for Society and Social Good, focusing on developing modeling and algorithmic foundations for tackling societal problems involving agents and predicting agent behavior in societal contexts, leveraging AI, game theory, mechanism design, and machine learning to better inform policymaking and (collective) decision-making. His team has been addressing societal challenges and fairness issues in various domains, including security (e.g., reducing vulnerability), public health (e.g., reducing substance use and homelessness), and urban planning (e.g., improving accessibility to public facilities), collaborating with domain experts. His research has been supported by NSF, NIH, and USCYBERCOM. He has received several Best Paper Awards at SDM and AAMAS and distinguished/outstanding SPC/PC member recognitions at IJCAI and WSDM. He has given tutorials and talks on computational game theory and mechanism design at venues such as AAMAS and IJCAI, including an Early Career Spotlight at IJCAI 2022. He has served as co-chairs for the AI and Social Good Track, Demonstration Track, Student Activities, Doctoral Consortium, Job Fair, Scholarships, Finance, and Diversity & Inclusion Activities at AAAI, AAMAS, and IJCAI.

Location: Old Computer Science, room 1310

Postmortem Program Analysis from a Conventional Program Analysis Method to an AI-assisted Approach

Abstract: Despite the best efforts of developers, software inevitably contains flaws that may be leveraged as security vulnerabilities. Modern operating systems integrate various security mechanisms to prevent software faults from being exploited. To bypass these defenses and hijack program execution, an attacker needs to constantly mutate an exploit and make many attempts. While in their attempts, the exploit triggers a security vulnerability and makes the running process abnormally terminate.

After a program has crashed and abnormally terminated, it typically leaves behind a snapshot of its crashing state in the form of a core dump. While a core dump carries a large amount of information, which has long been used for software debugging, it barely serves as informative debugging aids in locating software faults, particularly memory corruption vulnerabilities. As such, previous research mainly seeks fully reproducible execution tracing to identify software vulnerabilities in crashes. However, such techniques are usually impractical for complex programs. Even for simple programs, the overhead of fully reproducible tracing may only be acceptable at the time of in-house testing.

In this talk, I will discuss how we tackle this issue by bridging program analysis with artificial intelligence (AI). More specifically, I will first talk about the history of postmortem program analysis, characterizing and disclosing their limitations. Second, I will introduce how we design a new reverse-execution approach for postmortem program analysis. Third, I will discuss how we integrate AI into our reverse-execution method to escalate its analysis efficiency and accuracy. Last but not least, as part of this talk, I will demonstrate the effectiveness of this AI-assisted postmortem program analysis framework by using massive amounts of real-world programs.

Bio: Dr. Xinyu Xing is an Assistant Professor at Pennsylvania State University. His research interests include exploring, designing and developing new program analysis and AI techniques to automate vulnerability discovery, failure reproduction, vulnerability diagnosis (and triage), exploit and security patch generation. His past research has been featured by many mainstream media and received the best paper awards from ACM CCS and ACSAC. Going beyond academic research, he also actively participates and hosts many world-class cybersecurity competitions (such as HITB and XCTF). As the founder of JD-OMEGA, his team has been selected for DEFCON/GeekPwn AI challenge grand final at Las Vegas. Currently, his research is mainly supported by NSF, ONR, NSA and industry partners.
As artificial intelligence continues to transform higher education and the world beyond, how are students engaging with this change? Join us for a student-led discussion that explores how AI is influencing academic integrity, learning practices, and students' perspectives on its role in future workplaces.

Our panelists will share their experiences and reflections on questions such as:
1. What counts as appropriate and inappropriate use of AI in coursework?
2. How do faculty approach AI and talk about its implications in class?
3. What does AI mean for students' learning and ethical decision-making?
4. How are students building their understanding of AI tools and their potential uses in professional contexts?

This conversation offers an authentic look at how students are navigating the promises and challenges of AI--both in their studies and as they look ahead to applying these technologies responsibly in their fields.

Register here.