Thanks for sharing! Are there any positions that do not require a US work permit (or remote possibility)?

I'm excited about this line of work, as it paves the way towards more effective and lexically grounded representations for complex queries!
Many thanks to my co-authors for their support and constructive feedback!
#IR #NLP #research

To do so, we introduce a novel activation function (SNReLU) that preserves the desirable properties of splade models (log saturation effect, sparsity). We find that, given enough training data (needed to adapt to the new act fn), retrieval performance for negations improves by +100%!

Finally, we focus on negations, a challenging yet under-explored area in IR. The main idea is simple: we allow LSR models attribute negative term scores to terms associated with negated aspects of the query. In practice this allows LSR models to directly penalize documents containing negated aspects

First, we construct zero-shot representations
using vector arethmetics designed to encapsulate each compositionality type (eg. intersection or negations). This method often outperforms DR/LSR models trained for compositionality, which struggle for some of these compositionality types

I'd like to be added too!