G M Shahariar
2nd year Ph.D. Student
University of California, Riverside
About Me

I am a second year CS Ph.D. student in the Department of Computer Science and Engineering at the University of California - Riverside, where I spend most of my time thinking about (1) how to better understand and formalize the chain-of-thought (CoT) reasoning process of large reasoning models (LRMs), and (2) how to make multimodal LLMs (MLLMs) safer and more aligned with human values through mechanistic interpretability. I am very fortunate to be a member of the SEAS Lab and advised by Professor Nael Abu-Ghazaleh.

Summer 2026: I'll be working on agent orchestration as an applied science intern with the Agent 365 team at Microsoft Redmond under Vivian Sun. If you are also in Seattle/Redmond/Bellevue area, let's connect and chat .

I am also generally interested in computer vision, bias and fairness, low-resource and cross-lingual NLP, and AI for social good. If you are an undergraduate or masters student interested in working with me, reach me via email and send me a very short problem description you would like to work on.

Education
  • University of California, Riverside
    University of California, Riverside
    Ph.D. in Computer Science
    Sep 2024 - Present
  • University of California, Riverside
    University of California, Riverside
    M.Sc. in Computer Science
    Sep 2024 - Jun 2026
  • Ahsanullah University of Science and Technology
    Ahsanullah University of Science and Technology
    B.Sc. in Computer Science
    Sep 2014 - Dec 2018
Experience
  • Microsoft
    Microsoft
    Applied Science Intern
    (Team: Agent 365)
    Jun 2026 - Present
  • University of California, Riverside
    University of California, Riverside
    Graduate Research Assistant
    Graduate Teaching Assistant
    Sep 2024 - Present
  • BUET
    BUET
    Research Assistant
    Jan 2021 - Dec 2023
  • AUST
    AUST
    Lecturer
    Jul 2019 - Aug 2024
  • Enosis Solutions
    Enosis Solutions
    Software Engineer
    Oct 2018 - Dec 2018
Honors & Awards
  • ICML 2026 Gold Reviewer
    2026
  • ICML 2026 Spotlight + Oral
    2026
  • Dean’s Distinguished Fellowship Award (UCR)
    2024
  • Khan Bahadur Ahsanullah Medal (AUST)
    2023
  • Best Paper Award, International Conference on ECCE
    2023
  • Prime Minister Gold Medal (UGC, Bangladesh)
    2020
  • Dean’s List of Honor (AUST)
    2018
News
2026
Started working at Microsoft (Agent 365 Team) as an Applied Science Intern in Redmond, Washington. Internship
Jun 22
Received my M.Sc. from University of California, Riverside. Masters Graduation
Jun 12
Modeling Hierarchical Thinking in Large Reasoning Models is accepted at ICML 2026! Spotlight Oral Read More
May 23
Recognized as a Gold Reviewer by ICML 2026. Award
May 13
Are Vision Language Models Cross-Cultural Theory of Mind Reasoners? is accepted at ACL 2026 Workshop C3NLP! Read More
Apr 30
PII-VisBench is accepted at ACL 2026 Findings! Read More
Apr 05
Structural Input Perturbations Expose Multimodal Alignment Blind Spots is accepted at ICLR 2026! Poster Read More
Jan 26
Selected Publications (view all ) Google Scholar
Modeling Hierarchical Thinking in Large Reasoning Models
ICML
Modeling Hierarchical Thinking in Large Reasoning Models

G M Shahariar*; Erfan Shayegani*; Ali Nazari; Nael Abu-Ghazaleh. (* equal contribution )

ICML 2026 Oral

TL;DR: We model and steer LLM reasoning as outcome-aligned trajectories over FSM states, enabling sentence-level inference time control via latent interventions.

[Openreview] [Arxiv] [Code]

# LLM reasoning # chain-of-thought # FSM # mechanistic interpretability # activation steering

Modeling Hierarchical Thinking in Large Reasoning Models

G M Shahariar*; Erfan Shayegani*; Ali Nazari; Nael Abu-Ghazaleh. (* equal contribution )

ICML 2026 Oral

TL;DR: We model and steer LLM reasoning as outcome-aligned trajectories over FSM states, enabling sentence-level inference time control via latent interventions.

[Openreview] [Arxiv] [Code]

# LLM reasoning # chain-of-thought # FSM # mechanistic interpretability # activation steering

ICML
PII-VisBench: Evaluating Personally Identifiable Information Safety in Vision Language Models Along a Continuum of Visibility
ACL
PII-VisBench: Evaluating Personally Identifiable Information Safety in Vision Language Models Along a Continuum of Visibility

G M Shahariar; Zabir Al Nazi; Md Olid Hasan Bhuiyan; Zhouxing Shi.

ACL Findings 2026

TL;DR: We show that Vision Language Models leak more personal information about people with greater online visibility, introducing a benchmark that reveals visibility-dependent privacy failures.

[Openreview] [Arxiv] [Code]

PII-VisBench: Evaluating Personally Identifiable Information Safety in Vision Language Models Along a Continuum of Visibility

G M Shahariar; Zabir Al Nazi; Md Olid Hasan Bhuiyan; Zhouxing Shi.

ACL Findings 2026

TL;DR: We show that Vision Language Models leak more personal information about people with greater online visibility, introducing a benchmark that reveals visibility-dependent privacy failures.

[Openreview] [Arxiv] [Code]

ACL
Misaligned Roles, Misplaced Images: Structural Input Perturbations Expose Multimodal Alignment Blind Spots
ICLR
Misaligned Roles, Misplaced Images: Structural Input Perturbations Expose Multimodal Alignment Blind Spots

Erfan Shayegani*; G M Shahariar*; Sara Abdali; Lei Yu; Nael Abu-Ghazaleh; Yue Dong. (* equal contribution )

ICLR 2026 Poster

TL;DR: We model and steer LLM reasoning as outcome-aligned trajectories over FSM states, enabling sentence-level inference time control via latent interventions.

[Openreview] [Arxiv] [Code]

Misaligned Roles, Misplaced Images: Structural Input Perturbations Expose Multimodal Alignment Blind Spots

Erfan Shayegani*; G M Shahariar*; Sara Abdali; Lei Yu; Nael Abu-Ghazaleh; Yue Dong. (* equal contribution )

ICLR 2026 Poster

TL;DR: We model and steer LLM reasoning as outcome-aligned trajectories over FSM states, enabling sentence-level inference time control via latent interventions.

[Openreview] [Arxiv] [Code]

ICLR
Adversarial Attacks on Parts of Speech: An Empirical Study in Text-to-Image Generation
EMNLP
Adversarial Attacks on Parts of Speech: An Empirical Study in Text-to-Image Generation

G M Shahariar; Jia Chen; Jiachen Li; Yue Dong.

EMNLP Findings 2024

TL;DR: We investigate the impact of adversarial attacks on different POS tags within text prompts on the images generated by T2I models.

[Openreview] [ACL Anthology] [Code]

Adversarial Attacks on Parts of Speech: An Empirical Study in Text-to-Image Generation

G M Shahariar; Jia Chen; Jiachen Li; Yue Dong.

EMNLP Findings 2024

TL;DR: We investigate the impact of adversarial attacks on different POS tags within text prompts on the images generated by T2I models.

[Openreview] [ACL Anthology] [Code]

EMNLP
Bengali Fake Reviews: A Benchmark Dataset and Detection System
Neurocomputing
Bengali Fake Reviews: A Benchmark Dataset and Detection System

G M Shahariar*; Md. Tanvir Rouf Shawon*; Faisal Muhammad Shah; Mohammad Shafiul Alam; Md. Shahriar Mahbub. (* equal contribution )

Neurocomputing 2024

TL;DR: We introduce the Bengali Fake Review Detection (BFRD) dataset that focuses on food-related reviews in Bengali language.

[Elsevier] [Arxiv] [Code] [Dataset]

Bengali Fake Reviews: A Benchmark Dataset and Detection System

G M Shahariar*; Md. Tanvir Rouf Shawon*; Faisal Muhammad Shah; Mohammad Shafiul Alam; Md. Shahriar Mahbub. (* equal contribution )

Neurocomputing 2024

TL;DR: We introduce the Bengali Fake Review Detection (BFRD) dataset that focuses on food-related reviews in Bengali language.

[Elsevier] [Arxiv] [Code] [Dataset]

Neurocomputing
Contrastive Learning for API Aspect Analysis
ASE
Contrastive Learning for API Aspect Analysis

G M Shahariar; Tahmid Hasan; Anindya Iqbal; Gias Uddin.

ASE 2023

TL;DR: We present a novel approach - CLAA - for API aspect detection in API reviews that utilizes transformer models trained with a supervised contrastive loss objective function.

[IEEE Explore] [Arxiv] [Code]

Contrastive Learning for API Aspect Analysis

G M Shahariar; Tahmid Hasan; Anindya Iqbal; Gias Uddin.

ASE 2023

TL;DR: We present a novel approach - CLAA - for API aspect detection in API reviews that utilizes transformer models trained with a supervised contrastive loss objective function.

[IEEE Explore] [Arxiv] [Code]

ASE
All publications