transformers

It is an Electra model for Extractive Question Answering on Italian texts. I fine-tuned the pre-trained model dbmdz/electra-base-italian-xxl-cased-discriminator on SQuAD-it dataset. Strengths 🏋️‍ To date, the best publicly available model for Question Answering in Italian is it5/it5-large-question-answering by Gabriele Sarti and Malvina Nissim.

Starting in 2021, I got hooked on neural/semantic/vector search. A while back I realized that I’d built up some knowledge in this vibrant field and thought I’d start sharing some “Neural Search Pills” 💊!

This project aims to show that a naive and simple baseline for fact checking can be built by combining dense retrieval and a textual entailment task. In a nutshell, the flow is as follows:

Who killed Laura Palmer is a simple Question Answering system, based on data crawled from Twin Peaks Wiki. It is built using Haystack, an awesome open-source framework for building search systems that work intelligently over large document collections.

As part of the DANKMEMES competition (EVALITA 2020), I developed a model for multimodal detection of political memes. The proposed solution is a feed-forward neural network that leverages pretrained embeddings, both for text and image representation.