View Proposal

---

Proposer

Medhdhar Salem Ali Muthanna Al- Gaashani

Title

Zero-Shot Image Classification with Large Deep Learning Models

Goal

Learn and understand popular large deep-learning models and use them to classify natural images, adding extra learnable parameters to fine-tune pre-trained large models on a target dataset to achieve zero-shot image classification.

Description

This project explores how large pre-trained vision-language models can be adapted for zero-shot image classification. Students study the popular CLIP large deep-learning model and learn how to fine-tune deep neural networks on a target dataset, gaining an understanding of the supervised training mechanism for image classification. Rather than training a model from scratch, extra learnable parameters are added to the pre-trained CLIP model so it can be efficiently adapted to a new target dataset.

Resources

Radford, A., Kim, J. W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., Krueger, G., & Sutskever, I. (2021). Learning Transferable Visual Models From Natural Language Supervision. Proceedings of the 38th International Conference on Machine Learning (ICML), PMLR 139, pp. 8748–8763. Official PDF: https://proceedings.mlr.press/v139/radford21a/radford21a.pdf arXiv: https://arxiv.org/abs/2103.00020

Background

Url

Difficulty Level

High

Ethical Approval

None

Number Of Students

1

Supervisor

Medhdhar Salem Ali Muthanna Al- Gaashani

Keywords

llm, computer vision , zero-shot image classification

Degrees

Bachelor of Science in Computational Sciences and Software Engineering