Small Cues, Big Consequences: Learning Pivotal Cues for Multimodal Meme Classification
Accepted at the 2026 Conference on Empirical Methods in Natural Language Processing (EMNLP 2026, Findings)
AI Researcher & Engineer
Akshit Sharma
I work in the domain of AI Safety & trustworthy AI with earlier research work being in the domain of multimodal learning and misinformation analysis.
I am currently an SDE at Oracle, working on agentic AI and GenAI systems. I received my B.Tech in Computer Science and Engineering, with a minor in Data Science and Artificial Intelligence, from the Indian Institute of Technology Guwahati in 2025.
My research interests lie in AI safety, with prior research in multimodal misinformation and hateful content detection. My recent work includes papers accepted at EMNLP and CVPR workshops, alongside publications in other CORE A*/A-ranked venues. As of , I am available for volunteering, collaboration, and judging opportunities.
Recent news
Two of our papers have been accepted to the prestigious EMNLP (CORE A*)!
What Improves Multimodal Misinformation Detection? Answers from a Large-Scale Empirical Study
Accepted to the Tenth Widening NLP Workshop (WiNLP) at EMNLP 2026
“What Makes Multimodal Meme Classification Reliable? A Large-Scale Controlled Study of Fusion, Encoders, and Modality Reliance” has been accepted to the MULA Workshop at CVPR 2026.
Our paper “MemeTAG: Keyword-Driven Meme Classification through Tag Embedding Reconstruction” has been accepted to WACV 2026.
Graduated from IIT Guwahati with a B.Tech in Computer Science Engineering & a minor in Data Science & AI. Runner-up for the Dr. Shankar Dayal Sharma Gold Medal, awarded to a student from the graduating batch for overall excellence in academic performance, character, conduct, extracurricular activities, and social service.
Felicitated as the National Winner of the Goldman Sachs India Hackathon 2024 out of6,000+ students selected across India. [Link]
Secured the 1st position in the NetApp Cloud Hackathon out of 800+ shortlisted contestants!.
01
Selected publications
Selected peer-reviewed conference papers.
2026
Small Cues, Big Consequences: Learning Pivotal Cues for Multimodal Meme Classification
Accepted at The 2026 Conference on Empirical Methods in Natural Language Processing (EMNLP 2026, Findings)
2026
What Makes Multimodal Meme Classification Reliable? A Large-Scale Controlled Study of Fusion, Encoders, and Modality Reliance
Accepted at IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2026
2026
MemeTAG: Keyword-Driven Meme Classification through Tag Embedding Reconstruction
Accepted at IEEE/CVF Winter Conference on Applications of Computer Vision 2026 (WACV 26)
2025
Robust Fake News Detection via Adversarial Contrastive Learning over Vision–Language Representations
Accepted at 10th International Conference on Computer Vision and Image Processing 2025 (CVIP 25)
2024
HDL-SAM: A Hybrid Deep Learning Framework for High-Resolution Imaging in Scanning Acoustic Microscopy
Accepted at IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2024
02
Research experience
Appointments across multimodal learning and biomedical vision.
Indian Institute of Technology Guwahati
Graduate Research Assistantship under Dr. Prashant.W Patil · Guwahati, Assam (Hybrid)
Worked in the domain of multimodal misinformation analysis and explainable hateful meme classification with a special emphasis on enhancing the visual understanding of hateful memes. Recent works accepted at WACV 2025, CVPRW 2026 & EMNLP 2026.
Indian Institute of Technology, Guwahati
Undergraduate Researcher · Guwahati, India
Worked on applications of Machine Learning in the field of Remote Sensing (Supervised by Dr. Arijit Sur). Skills: Deep Learning, Computer Vision, Python.
Harvard University
Research Internship
Part-time research internship at Harvard University.
UiT The Arctic University of Norway
Research Assistantship, under Dr. Anowarul Habib · Tromsø, Norway (Remote)
Worked on hybrid deep learning for Scanning Acoustic Microscopy (HDL-SAM), designing a two-stage hypergraph + transformer pipeline that significantly improved reconstruction quality over existing super-resolution and inpainting methods while reducing the need for slow, dense acoustic scanning. Work accepted at SynData4CV CVPR Workshop.
Chubu University, Japan
Research Internship, supervised by Dr. Yuji Iwahori · Kasugai, Japan (Remote)
Worked on super-resolution in colonoscopy polyp analysis under Dr. Yuji Iwahori, developing a novel parameter-efficient model, giving competitive performance in polyp classification while introducing innovative enhancement techniques.
Georgia Institute of Technology
Research Internship · Georgia, United States · Remote
Actively worked in the field of synthesis of reactive systems (particularly LTLf model synthesis) – devised and implemented an algorithm to check satisfiability of a given LTLf formula under a terminating system and worked toward an on-the-fly algorithm to greatly reduce memory requirements.
MBZUAI (Mohamed bin Zayed University of Artificial Intelligence)
Research Internship, under Dr. Hisham Cholakkal · Abu Dhabi, UAE (Remote)
Worked in computer vision for biomedical image segmentation, implementing UNet and Swin-UNet architectures for cellular instance segmentation, and developing a model ensemble combining SAM and Cellpose to enhance prediction accuracy.
03
Professional experience
Applied machine learning and software systems.
Oracle India Private Limited
Member of Technical Staff · Bengaluru
Developing language-model features for service-request analysis, including summarization, tag generation, clustering and conversational analytics. My work also includes evaluation and safeguards for deployed model endpoints.
Oracle India Private Limited
Software Engineering Intern · Bengaluru
Worked on authentication, authorization, token lifecycle management and service initialization for an internal Kubernetes cluster-management platform.
04
Selected projects
Research implementations and software systems.
Agentic Red-Teaming Framework for LLM Safety Evaluation
An automated framework for adversarial attack generation, multi-turn testing, structured reporting and model-based evaluation of LLM safeguards.
LatentMotion: GAN-based Animation from a Single Image
A StyleGAN2-based animation system that inverts a single image into the W/W+ latent space and generates coherent video frames through learned semantic directions for pose, expression and style variations.
DQN-Based Maze Solving Agent for Cumulative Reward Maximization
A Deep Q-Learning agent that learns optimal maze policies with a neural-network Q-function approximator, without relying on a tabular Q-table.
05
Education & honors
Academic background and selected recognition.
Education
Indian Institute of Technology Guwahati
B.Tech in Computer Science & Engineering, 2025
Minor in Data Science & Artificial Intelligence
Selected honors
- National Winner, Goldman Sachs India Hackathon 2024
- 1st place, NetApp Cloud Hackathon at IIT Guwahati
- Runner-up, Dr. Shankar Dayal Sharma Gold Medal
- Winner, Adobe PapyrusNebulae Hackathon 2023
- International Rank 2 (Silver Medallist) and International Rank 6, International Computer Science Olympiad, 2016 and 2017
- Kaggle Grandmaster