Daniel Simig

Daniel Simig, Fabio Petroni, Pouya Yanki et al. · meta-ai

The extreme multi-label classification (XMC) task aims at tagging content with a subset of labels from an extremely large label set. The label vocabulary is typically defined in advance by domain experts and assumed to capture all necessary tags. However in real world scenarios this label set, although large, is often incomplete and experts frequently need to refine it. To develop systems that simplify this process, we introduce the task of open vocabulary XMC (OXMC): given a piece of content, predict a set of labels, some of which may be outside of the known tag set. Hence, in addition to not having training data for some labels - as is the case in zero-shot classification - models need to invent some labels on-the-fly. We propose GROOV, a fine-tuned seq2seq model for OXMC that generates the set of labels as a flat sequence and is trained using a novel loss independent of predicted label order. We show the efficacy of the approach, experimenting with popular XMC datasets for which GROOV is able to predict meaningful labels outside the given vocabulary while performing on par with state-of-the-art solutions for known labels.

Kushal Tirumala, Daniel Simig, Armen Aghajanyan et al.

Over recent years, an increasing amount of compute and data has been poured into training large language models (LLMs), usually by doing one-pass learning on as many tokens as possible randomly selected from large-scale web corpora. While training on ever-larger portions of the internet leads to consistent performance improvements, the size of these improvements diminishes with scale, and there has been little work exploring the effect of data selection on pre-training and downstream performance beyond simple de-duplication methods such as MinHash. Here, we show that careful data selection (on top of de-duplicated data) via pre-trained model embeddings can speed up training (20% efficiency gains) and improves average downstream accuracy on 16 NLP tasks (up to 2%) at the 6.7B model scale. Furthermore, we show that repeating data intelligently consistently outperforms baseline training (while repeating random data performs worse than baseline training). Our results indicate that clever data selection can significantly improve LLM pre-training, calls into question the common practice of training for a single epoch on as much data as possible, and demonstrates a path to keep improving our models past the limits of randomly sampling web data.

Sebastian Cadavid-Sanchez, Khalil Kacem, Rafael Aparecido Martins Frade et al.

To study social, economic, and historical questions, researchers in the social sciences and humanities have started to use increasingly large unstructured textual datasets. While recent advances in NLP provide many tools to efficiently process such data, most existing approaches rely on generic solutions whose performance and suitability for domain-specific tasks is not well understood. This work presents an attempt to bridge this domain gap by exploring the use of modern Entity Linking approaches for the enrichment of museum collection data. We collect a dataset comprising of more than 1700 texts annotated with 7,510 mention-entity pairs, evaluate some off-the-shelf solutions in detail using this dataset and finally fine-tune a recent end-to-end EL model on this data. We show that our fine-tuned model significantly outperforms other approaches currently available in this domain and present a proof-of-concept use case of this model. We release our dataset and our best model.