Search for: All records

The eXtreme Multi-label Classification (XMC) problem seeks to find relevant labels from an exceptionally large label space. Most of the existing XMC learners focus on the extraction of semantic features from input query text. However, conventional XMC studies usually neglect the side information of instances and labels, which can be of use in many real-world applications such as recommendation systems and e-commerce product search. We propose Predicted Instance Neighborhood Aggregation (PINA), a data enhancement method for the general XMC problem that leverages beneficial side information. Unlike most existing XMC frameworks that treat labels and input instances as featureless indicators and independent entries, PINA extracts information from the label metadata and the correlations among training instances. Extensive experimental results demonstrate the consistent gain of PINA on various XMC tasks compared to the state-of-the-art methods: PINA offers a gain in accuracy compared to standard XR-Transformers on five public benchmark datasets. Moreover, PINA achieves a ∼ 5% gain in accuracy on the largest dataset LF-AmazonTitles-1.3M. Our implementation is publicly available https://github.com/amzn/pecos/ tree/mainline/examples/pina. 

We are concerned with the problem of designing large families of subsets over a common labeled ground set that have small pairwise intersections and the property that the maximum discrepancy of the label values within each of the sets is less than or equal to one. Our results, based on transversal designs, factorizations of packings and Latin rectangles, show that by jointly constructing the sets and labeling scheme, one can achieve optimal family sizes for many parameter choices. Probabilistic arguments akin to those used for pseudorandom generators lead to significantly suboptimal results when compared to the proposed combinatorial methods. The design problem considered is motivated by applications in molecular data storage.