Projects

Multilingual Article Generation from Knowledge Bases

In this project, we aim to generate millions of Wikipedia articles in various languages from triples extracted from knowledge base sources like Wikidata. Our generation models work as a loop, extracting triples from texts, then generating texts from triples, and so on to update frequently the new data. Next, we apply pre-defined conditions and metrics to validate the generated texts compared to inputs to improve their quality. Finally, the generated texts are fed to the evaluation model, which merges human evaluation to guarantee the best outcome quality. We use large language models (ChatGPT, GPT-4, Llama, FLAN, etc.) and optimization techniques to work best with downstream tasks.

Kolmogorov-Arnold networks

Check my GitHub: https://github.com/hoangthangta for more details.

BSRBF-KAN: https://github.com/hoangthangta/BSRBF_KAN/blob/main/BSRBF-KAN.pdf

https://github.com/hoangthangta/BSRBF_KAN/blob/main/BSRBF-KAN.pptx