Model Compression and AutoML for Efficient Click-Through Rate Prediction
Novel machine learning architectures can adeptly learn to predict user response for recommender systems. However, these model architectures are often effective at the cost of large computational, and memory, cost. This limits their ability to run on edge devices with smaller hardwares, such as smart...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Published: |
Massachusetts Institute of Technology
2022
|
| Online Access: | https://hdl.handle.net/1721.1/139253 |
| _version_ | 1826207356269101056 |
|---|---|
| author | Gschwind, Katharina |
| author2 | Han, Song |
| author_facet | Han, Song Gschwind, Katharina |
| author_sort | Gschwind, Katharina |
| collection | MIT |
| description | Novel machine learning architectures can adeptly learn to predict user response for recommender systems. However, these model architectures are often effective at the cost of large computational, and memory, cost. This limits their ability to run on edge devices with smaller hardwares, such as smartphones, which is a popular use case for recommender systems. We address this issue in this thesis by studying how compression of recommender system models can significantly reduce model computation cost, and edge device runtime, while preserving prediction accuracy. Furthermore, we present a new compression-based AutoML method for feature set generation in architectures which incorporate explicit feature interactions. This works as a tool to build efficient recommender system models, and is applicable to many state of the art model designs. Applying this AutoML shows initial gains in model performance. |
| first_indexed | 2024-09-23T13:48:03Z |
| format | Thesis |
| id | mit-1721.1/139253 |
| institution | Massachusetts Institute of Technology |
| last_indexed | 2024-09-23T13:48:03Z |
| publishDate | 2022 |
| publisher | Massachusetts Institute of Technology |
| record_format | dspace |
| spelling | mit-1721.1/1392532022-01-15T03:25:12Z Model Compression and AutoML for Efficient Click-Through Rate Prediction Gschwind, Katharina Han, Song Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science Novel machine learning architectures can adeptly learn to predict user response for recommender systems. However, these model architectures are often effective at the cost of large computational, and memory, cost. This limits their ability to run on edge devices with smaller hardwares, such as smartphones, which is a popular use case for recommender systems. We address this issue in this thesis by studying how compression of recommender system models can significantly reduce model computation cost, and edge device runtime, while preserving prediction accuracy. Furthermore, we present a new compression-based AutoML method for feature set generation in architectures which incorporate explicit feature interactions. This works as a tool to build efficient recommender system models, and is applicable to many state of the art model designs. Applying this AutoML shows initial gains in model performance. M.Eng. 2022-01-14T14:59:39Z 2022-01-14T14:59:39Z 2021-06 2021-06-17T20:13:15.997Z Thesis https://hdl.handle.net/1721.1/139253 In Copyright - Educational Use Permitted Copyright MIT http://rightsstatements.org/page/InC-EDU/1.0/ application/pdf Massachusetts Institute of Technology |
| spellingShingle | Gschwind, Katharina Model Compression and AutoML for Efficient Click-Through Rate Prediction |
| title | Model Compression and AutoML for Efficient Click-Through Rate Prediction |
| title_full | Model Compression and AutoML for Efficient Click-Through Rate Prediction |
| title_fullStr | Model Compression and AutoML for Efficient Click-Through Rate Prediction |
| title_full_unstemmed | Model Compression and AutoML for Efficient Click-Through Rate Prediction |
| title_short | Model Compression and AutoML for Efficient Click-Through Rate Prediction |
| title_sort | model compression and automl for efficient click through rate prediction |
| url | https://hdl.handle.net/1721.1/139253 |
| work_keys_str_mv | AT gschwindkatharina modelcompressionandautomlforefficientclickthroughrateprediction |