DeepSketch: A New Machine Learning-Based Reference Search Technique...

on YouTube ↗15m

" summary

A machine-learned reference search called DeepSketch extracts binary sketches via a small DNN (three conv + two FC layers) and uses clustering plus transfer learning to map similar blocks for post-deduplication delta compression. It builds a dynamic k-means clustering pipeline and resized clusters for balanced training, then applies learning-to-hash; evaluation on eleven workloads shows up to 33% improvement (21% average) in data-reduction over a state-of-the-art baseline, with measurable runtime overhead.

AI-written summary · about

" chapters (15)

1. 0:00Intro
2. 0:29Executive Summary
3. 2:08Data Reduction in Storage Systems
4. 2:25Post-deduplication Delta Compression Combines three different data-reduction approaches
5. 2:47Overview of Post-Deduplication Delta Compression
6. 3:54Lossless Compression
7. 4:05Key Challenge: Reference Search How to find a good reference block for an incoming data block across a wide range of stored data at low cost
8. 5:38Limitations of Existing Techniques - Provide significantly lower data-reduction ratios than the optimal
9. 6:31DeepSketch: Key Idea Use the learning-to-hash method for sketch generation A promising machine learning (ML).-based approach for the
10. 7:22DeepSketch: Challenges Lack of semantic information
11. 8:32Data Clustering for DeepSketch . Existing clustering algorithms are unsuitable for DeepSketch
12. 9:44Post-Processing for Training Data Set Non-uniform distribution of data blocks across the clusters
13. 11:09Evaluation Methodology Compared data-reduction techniques
14. 11:52Overall Data-Reduction Benefits
15. 13:25Performance Overhead

" related talks

• 06m#ProjectDeepFillJai way, Jeff Wayne · Adobe MAX 2017
• 19mSketchforme: Composing Sketched Scenes from Text Descriptions for Interactive ApplicationsForrest Huang, John F. Canny · UIST 2019
• 31mDoug Eck, Research Scientist, GoogleDoug Eck · MLconf SF 2017
• 43mThe Search for a New Visual Search, Beyond LanguageBryan Minor · Strange Loop 2018
• 37mSimilarity Search for Product Matching @ Semantics3Abishek Bhat · The Fifth Elephant 2019
• 37mJavaScript Machine Learning from Magic to ScienceGant Laborde · Devnexus 2021
• 01h 10mSketching for Linear Algebra: Basics of Dimensionality Reduction and CountSketch IIDavid P. Woodruff · Foundations of Data Science Boot Camp 2018
• 27mDeepCS: a Code Search Tool Powered by Deep LearningRobert Rodger · Berlin Buzzwords 2019