Abhijit Bhattacharjee / MathWorks Inc.

機器學習與深度學習在各產業間被廣泛視為解決複雜建模問題的強大工具。採用機器學習的好處可見於各種應用，包含預測性維護、健康監控、財經投資組合預測以及先進駕駛輔助等。

然而，想從訊號資料開發出預測模型卻不是一項簡單的任務。此外，近來對於追求更聰明、更有智慧的感測器訊號處理演算法的開發需求正急速升溫，並將這些演算法佈署到物聯網端點裝置或雲端之上。

請參加這段演講，幫助你了解更多關於訊號處理和小波的最新功能，讓你更容易地透過感測器資料執行機器學習與深度學習。利用真實資料，我們將探索幾個訊號分類工作流程。演講內容囊括:

• 為訊號分類自動擷取特徵
• 以感測器資料透過CNN與LSTM網路進行遷移式學習的工作流程，開發出預測模型
• 訊號前處理技巧以提升訊號品質
• 透過tall arrays等新資料類型來處理無法容納於記憶體的資料，減輕撰寫特殊程式碼來處理龐大感測器資料的負擔。
• 利用多核心電腦、GPU、電腦叢集等高效能運算資源來提升表現

Machine learning and deep learning are powerful tools for solving complex modeling problems across a broad range of industries. The benefits of machine learning are being realized in applications everywhere, including predictive maintenance, health monitoring, financial portfolio forecasting, and advanced driver assistance to name a few applications.

However, identifying and extracting the relevant features for developing predictive models for sensor data is not a trivial task. Moreover, there is an increasing need for developing smart sensor signal processing algorithms which can be either deployed on edge nodes / embedded devices or on the cloud depending on the application.

Join us to learn more about the latest capabilities in signal processing and wavelets enable to you to perform machine learning and deep learning on sensor data with great ease. Using real data, we will explore a couple of workflows for signal classification:

Topics Include:

• Automatic Feature Extraction for signal classification
• Transfer learning workflows to develop predictive models on sensor data using CNN and LSTM networks
• Signal Pre-processing techniques to increase the signal quality
• New datatypes such as tall arrays to work with data that does not fit in memory alleviating the need for writing special code to work with large sensor data,
• Leverage high-performance computing resources, such as multicore computers, GPUs, computer clusters to scale up the performance