NIPS 2018
Sun Dec 2nd through Sat the 8th, 2018 at Palais des Congrès de Montréal

### Reviewer 1

- This paper proposes a multi-instance learning formulation combined with an early prediction technique to address the challenge of fast and efficient classification of sequential data on tiny/resource-constrained devices. The main driving observations include (a) presence of a signature in a small fraction of the data, and (b) that the signatures are predictable using a prefix of the data. The algorithm, EMI-RNN, Early Multi-Instance RNN, shows significant reduction in computation (about 80%) while maintaining/improving the accuracy of the models marginally. - The proposed algorithms E-RNN, and EMI-RNN are based on insights that are described intuitively in the paper. - The gains are impressive and the analysis of the proposed algorithms helps establish usefulness of the approach in real-world deployments of the models. - How sensitive is EMI-RNN to amount of noise? Discussion on gains achieved (accuracy, resources used, and time taken) with respect to percentage of noise in the data should help establish the importance of the approach taken by EMI-RNN. - a thorough analysis of the results with respect to different values of the parameter, $p^{hat}$, added to the appendix, should strengthen the evaluation. - The paper reads well, but another finishing pass should get rid of minor typos: - e.g., line 24 “a wrist bands” —> either “wrist bands” or “a wrist band”. - line 146: “that that” —> that - missing a period, line 118, between “updates” and “They”. - line 143-144-145 doesn’t read well, needs rephrasing. - Notation used for Equation 3.1.1. is explained after an entire paragraph (in a paragraph starting at line 18). This makes reader uncomfortable and causes doubts. Please consider restructuring this part. - Line 281 mentions that Table 2 is provided in the supplement. I understand this decision was taken due to space limitation, but this data is required to establish the appropriateness of the evaluation section. This needs to be included in the main manuscript. - Also, was the conclusions section removed in the interest of space? Please include a concise summary of findings, and next steps. ---------------------- I want to thank the authors for answering my questions; the response helped. I am more convinced of my decision of accepting the paper given that the authors are planning to address the comments in the final version.