We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.LG

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Machine Learning

Title: From Multimodal to Unimodal Attention in Transformers using Knowledge Distillation

Abstract: Multimodal Deep Learning has garnered much interest, and transformers have triggered novel approaches, thanks to the cross-attention mechanism. Here we propose an approach to deal with two key existing challenges: the high computational resource demanded and the issue of missing modalities. We introduce for the first time the concept of knowledge distillation in transformers to use only one modality at inference time. We report a full study analyzing multiple student-teacher configurations, levels at which distillation is applied, and different methodologies. With the best configuration, we improved the state-of-the-art accuracy by 3%, we reduced the number of parameters by 2.5 times and the inference time by 22%. Such performance-computation tradeoff can be exploited in many applications and we aim at opening a new research area where the deployment of complex models with limited resources is demanded.
Comments: Preprint. Final paper accepted at the 17th IEEE International Conference on Advanced Video and Signal-based Surveillance, AVSS 2021, Virtual, November 16-19, 2021. 10 pages
Subjects: Machine Learning (cs.LG); Computation and Language (cs.CL)
Cite as: arXiv:2110.08270 [cs.LG]
  (or arXiv:2110.08270v2 [cs.LG] for this version)

Submission history

From: Dhruv Agarwal [view email]
[v1] Fri, 15 Oct 2021 12:30:21 GMT (1773kb,D)
[v2] Tue, 19 Oct 2021 04:55:27 GMT (1777kb,D)

Link back to: arXiv, form interface, contact.