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

Download:

Current browse context:

cs.CL

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 > Computation and Language

Title: Exploring Low-dimensional Intrinsic Task Subspace via Prompt Tuning

Abstract: How can pre-trained language models (PLMs) learn universal representations and effectively adapt to broad NLP tasks differing a lot superficially? In this work, we empirically find evidences indicating that the adaptations of PLMs to various tasks can be reparameterized as optimizing only a few free parameters in a common low-dimensional intrinsic task subspace, which may help us understand why PLMs could easily adapt to various NLP tasks with small-scale data. Specifically, to find such a subspace and examine its universality, we resort to the recent success of prompt tuning and decompose the soft prompts of multiple NLP tasks into the same low-dimensional nonlinear subspace, then we learn to adapt the PLM to unseen tasks or data by only tuning parameters in the subspace. We dub this pipeline as intrinsic prompt tuning (IPT). In experiments, we study diverse few-shot NLP tasks and surprisingly find that in a 5-dimensional subspace found with 100 random tasks, by only tuning 5 free parameters, we can recover 87% and 65% of the full prompt tuning performance for 100 seen tasks (using different training data) and 20 unseen tasks, respectively, showing great generalization ability of the found intrinsic task subspace. Besides being an analysis tool, IPT could further bring practical benefits, such as improving the prompt tuning stability.
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
Cite as: arXiv:2110.07867 [cs.CL]
  (or arXiv:2110.07867v1 [cs.CL] for this version)

Submission history

From: Yujia Qin [view email]
[v1] Fri, 15 Oct 2021 05:43:59 GMT (313kb,D)

Link back to: arXiv, form interface, contact.