Lightweight Transformers for Zero-Shot and Fine-Tuned Text-to-SQL Generation Using Spider

Text-to-SQL translation enables non-expert users to query relational databases using natural language, with applications in education and business intelligence. This study evaluates three lightweight transformer models - T5-Small, BART-Small, and GPT-2 - on the Spider dataset, focusing on low-resource settings. We developed a reusable, model-agnostic pipeline that tailors schema formatting to each model's architecture, training them across 1000 to 5000 iterations and evaluating on 1000 test samples using Logical Form Accuracy (LFAcc), BLEU, and Exact Match (EM) metrics. Fine-tuned T5-Small achieves the highest LFAcc (27.8%), outperforming BART-Small (23.98%) and GPT-2 (20.1%), highlighting encoder-decoder models' superiority in schema-aware SQL generation. Despite resource constraints limiting performance, our pipeline's modularity supports future enhancements, such as advanced schema linking or alternative base models. This work underscores the potential of compact transformers for accessible text-to-SQL solutions in resource-scarce environments.

A deep learning approach to track eye movements based on events

This research project addresses the challenge of accurately tracking eye movements during specific events by leveraging previous research. Given the rapid movements of human eyes, which can reach speeds of 300{\deg}/s, precise eye tracking typically requires expensive and high-speed cameras. Our primary objective is to locate the eye center position (x, y) using inputs from an event camera. Eye movement analysis has extensive applications in consumer electronics, especially in VR and AR product development. Therefore, our ultimate goal is to develop an interpretable and cost-effective algorithm using deep learning methods to predict human attention, thereby improving device comfort and enhancing overall user experience. To achieve this goal, we explored various approaches, with the CNN\_LSTM model proving most effective, achieving approximately 81\% accuracy. Additionally, we propose future work focusing on Layer-wise Relevance Propagation (LRP) to further enhance the model's interpretability and predictive performance.