Presentation for interview
Download as PPTX, PDF0 likes16 views
This document provides biographical information about Shih Huan and summarizes his academic and research experience. It outlines his educational background, including undergraduate and graduate studies. It also describes his military service and a research project on gait recognition during that time. Finally, it summarizes his dissertation research on compressed learning for time series classification.
Presentation for interview
- 11. Data collection Passive user identification 13Gait Recognition
- 14. • Editor • Draft • Drill • Account 16Social Engineering
- 15. Compressed Learning for Time Series Classification 17 論文研究
- 16. 20Compressed sensing ► Most real-world signals are sparse in some basis A𝑥 = 𝑦, A ∈ ℝ 𝑝×𝑛 𝑎𝑛𝑑 𝑝 ≪ 𝑛 ► Dramatically reduce the transmission loading a measure
- 17. 21Compressed sensing Mostafa Mohsenvand Projects ► 𝑥 should be a 𝑘-sparse signal » 1 to 1 relation between data and compressed domain • A = 𝑟𝑎𝑛𝑑𝑛 𝑝,𝑛 𝑛 (mean=0, 𝜎 = 1 𝑛 )
- 18. 23Sparse representation http://www.investopedia.com/ ► The ‘envelope’ in finance
- 19. 24Sparse representation ► Given 𝑫, envelope with size 𝒌 𝑬 𝒌 = 𝒁 𝒁 = 𝒛 𝟏, 𝒛 𝟐, … 𝒛 𝒏 , 𝒛𝒋 − 𝝁𝒋 ≤ 𝒌 ∙ 𝒔𝒕𝒅𝒋 , ∀ 𝒛𝒋 ∈ ℝ} 𝝁𝒋 = 𝒎𝒆𝒂𝒏(𝑻𝒋) , 𝒔𝒕𝒅𝒋 = 𝒔𝒕𝒅(𝑻𝒋) ► Profiling the time series dataset
- 20. 25Sparse representation ► Encoding time series T as a sparse series S 𝑠𝑗 = 1, 𝑖𝑓 𝑡𝑗 > 𝜇 𝑗 + 𝑘 ∙ 𝑠𝑡𝑑𝑗 𝑠𝑗 = −1, 𝑖𝑓 𝑡𝑗 < 𝜇 𝑗 − 𝑘 ∙ 𝑠𝑡𝑑𝑗 𝑠𝑗 = 0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒 , 𝑓𝑜𝑟 𝑗 = 1 𝑡𝑜 𝑛 ► Sparsity indicates similarity between a time series and 𝐷
- 21. 28Sparse representation ► Sparsity indicates similarity ► Encoding result is interpretable
- 22. 29Workflow ► From raw series to feature
- 23. 31Experimental results ► Compression ratio = 𝑝/𝑛 (Number of measurements) / (data dimension)
- 24. Any questions? You can find me at: » shih0916@outlook.com 33THANKS! http://www.freeimageslive.co.uk/
- 25. CREDITS Special thanks to all the people who made and released these awesome resources for free: » Presentation template by SlidesCarnival » Photographs by Unsplash » Diverse device hand photos by Facebook Design Resources 34
- 26. PRESENTATION DESIGN This presentation uses the following typographies and colors: » Titles: Montserrat » Body copy: Source Sans Pro You can download the fonts on these pages: https://www.fontsquirrel.com/fonts/montserrat https://www.fontsquirrel.com/fonts/source-sans-pro » Sky Blue #00bef2 / Navy Blue #25516c You don’t need to keep this slide in your presentation. It’s only here to serve you as a design guide if you need to create new slides or download the fonts to edit the presentation in PowerPoint® 35
Editor's Notes
- #3: Outline Personal profile & my research
- #5: Master degree in computer science contact information
- #6: Personal experience School days Military life
- #7: 開發此應用的靈感來自於學校舉辦的多場演講。演講中，偶爾總會有人忘記將手機鈴聲設定成震動或靜音，導致鈴聲大作的尷尬場面。在"如果能有一個APP能夠自動且準確地管理好手機提醒之模式的話，這個問題便能夠被解決"的想法下，我們決定製作Smart Ring APP。 本軟體的主要功能就是提供使用者個人化的手機模式設定(目前以來電通知為主，並可再加入飛航模式、藍芽、WIFI、3G行動網路、音樂\遊戲音量、螢幕亮度、螢幕休眠時間等。) ，主要分為兩個部份
- #8: 長按後會跳出選單，選擇模式 時間到達後模式即切換 事件結束後會切換回原模式
- #9: 再來是點選地點，進入設定畫面 可透過網路直接擷取當前地址
- #10: 同上述方式，為其設定模式 需要幾秒的辨識時間，符合條件即切換
- #11: 希望能帶給人們健康又舒適的家居生活。我們研發智慧型家居系統，使用者可以下載模組，讓系統學習並辨識家中成員的行為模式。
- #12: Sketch may of gait cycle
- #13: EcoBT Mini 33Hz Using BLE for transmission
- #14: Less intrusiveness way to perform user identification
- #15: Intel DEMO day in Song-Shan cultural park
- #16: 利用大衆的疏於防範的小詭計，讓受害者掉入陷阱。該技巧通常以交談、欺騙、假冒或口語用字等方式，從合法用戶中套取用戶系統的秘密，例如：用戶名單、用戶密碼及網絡結構
- #18: Thesis topic Compression scheme Time series representation mehtod Classification framework
- #19: Data collected continuously, time correlated series; communication & storage issue Reduce dimension & recover with little loss Extract info. From TS for ML model
- #20: Lower transmission loading & recover well 取樣率 (f s) > 2 *受測訊號的最高頻率部份, 否則高頻的內容會失真(Aliasing)
- #21: The goal of compressed sensing is to provide measurement matrix A, with the number of measurements m as small as possible M is the #sample, which is << Nyquist
- #22: Normal Random matrix A generated with specific parameters is usually good enough for most real world applications.
- #23: Lower transmission loading & recover well 取樣率 (f s) > 2 *受測訊號的最高頻率部份, 否則高頻的內容會失真(Aliasing)
- #24: Exists in finance field 開盤價(opening)和收盤價(closing) identify extreme overbought and oversold
- #25: Definition of envelope with size k: any time series Z in dataset D
- #26: The choice of k is critical
- #27: 兩端分別是由開盤價和收盤價 identify extreme overbought and oversold
- #28: Propose a heuristic to make envelope distinguishable k --- tradeoff between sparsity and distinguishability X: size of envelope, Y: nonzero ratio of encoding result
- #29: Sparse property ⟹ transmission efficiency Encode data from same class get more sparse result With support of domain knowledge expert
- #30: Mainly focus on compressed represnetation
- #31: Concatenate all encoding result from each class
- #32: Keep good distinguishability with low compression ratio
- #33: Demo video using the method we propose