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: Making Good on LSTMs' Unfulfilled Promise

Abstract: LSTMs promise much to financial time-series analysis, temporal and cross-sectional inference, but we find that they do not deliver in a real-world financial management task. We examine an alternative called Continual Learning (CL), a memory-augmented approach, which can provide transparent explanations, i.e. which memory did what and when. This work has implications for many financial applications including credit, time-varying fairness in decision making and more. We make three important new observations. Firstly, as well as being more explainable, time-series CL approaches outperform LSTMs as well as a simple sliding window learner using feed-forward neural networks (FFNN). Secondly, we show that CL based on a sliding window learner (FFNN) is more effective than CL based on a sequential learner (LSTM). Thirdly, we examine how real-world, time-series noise impacts several similarity approaches used in CL memory addressing. We provide these insights using an approach called Continual Learning Augmentation (CLA) tested on a complex real-world problem, emerging market equities investment decision making. CLA provides a test-bed as it can be based on different types of time-series learners, allowing testing of LSTM and FFNN learners side by side. CLA is also used to test several distance approaches used in a memory recall-gate: Euclidean distance (ED), dynamic time warping (DTW), auto-encoders (AE) and a novel hybrid approach, warp-AE. We find that ED under-performs DTW and AE but warp-AE shows the best overall performance in a real-world financial task.
Comments: 33rd Conference on Neural Information Processing Systems (NeurIPS 2019), Vancouver, Canada. arXiv admin note: text overlap with arXiv:1812.02340
Subjects: Machine Learning (cs.LG); Computational Finance (q-fin.CP); Portfolio Management (q-fin.PM); Machine Learning (stat.ML)
Cite as: arXiv:1911.04489 [cs.LG]
  (or arXiv:1911.04489v4 [cs.LG] for this version)

Submission history

From: Daniel Philps [view email]
[v1] Mon, 11 Nov 2019 18:57:37 GMT (341kb,D)
[v2] Sat, 23 Nov 2019 14:56:54 GMT (316kb,D)
[v3] Sun, 1 Dec 2019 12:44:27 GMT (347kb,D)
[v4] Mon, 9 Dec 2019 02:07:27 GMT (375kb,D)

Link back to: arXiv, form interface, contact.