Open Edge AI & TinyML

Open Edge AI & TinyML

議程簡介

近年來隨著軟韌硬體的飛快進步及各項AI算法的到位，各項智能應用如雨後春筍般快速掘起，過去像智慧物聯網(AIoT)等相關應用都還需仰賴雲端來提供AI運算服務，如今在完全離線下亦可以實現，像是語音喚醒、運動偵測、異常偵測、影像分類、物件偵測等應用。此次議程希望召集更多有志一同的伙伴來分享一下關於開源離網邊緣智能（Edge AI）及微型機器學習(TinyML)的成果，期待讓更多人了解這項技術並落地，以便擁有更方便的生活。

Recently, with the rapid progress of software, hardware and various AI algorithms, various intelligent applications have sprung up like mushrooms after rain. In the past, related applications such as AIoT still relied on cloud to provide AI computing services. Now it can be achieved completely offline. Applications such as voice wake-up, motion detection, anomaly detection, image classification, object detection and other applications can be realized. This session hopes to gather more like-minded partners to share their achievements in open source Edge AI (Edge intelligence) and TinyML (Tiny Machine Learning & AI). We hope to let more people understand this technology and land it so that they can have a more convenient life.

篩選條件

議題

TinyML with MicroPython on Raspberry Pi Pico．半套

TR 615 [[ new Date( '2023-07-29 02:00:00+00:00' ).toLocaleDateString('ja', {year: 'numeric', month: '2-digit', day: '2-digit'}) ]] [[ new Date( '2023-07-29 02:00:00+00:00' ).toLocaleTimeString('zh-Hant', {hour12: false, hour: '2-digit', minute:'2-digit'}) ]] ~ [[ new Date( '2023-07-29 02:30:00+00:00' ).toLocaleTimeString('zh-Hant', {hour12: false, hour: '2-digit', minute:'2-digit'}) ]] zh-tw
加入行事曆 加入關注 加入關注 已關注

AIoT 時代，如何讓微控制器等級的邊緣裝置做做數據分析？

我們將在本場次說明 * 從 IoT 到 AIoT * tinyML 介紹 * MicroPython 介紹 * 使用 MicroPython 收資料後以 tinyML 在 Raspberry Pi Pico 推論 [DEMO] * 學習資源

講者

sosorry

sosorry

Hi, I'm sosorry.

Open Edge AI & TinyML GUHJYR general (30mins)

TOSA: 用於深度神經網路中張量操作 (tensor operations) 的嶄新開源架構

TR 615 [[ new Date( '2023-07-29 02:40:00+00:00' ).toLocaleDateString('ja', {year: 'numeric', month: '2-digit', day: '2-digit'}) ]] [[ new Date( '2023-07-29 02:40:00+00:00' ).toLocaleTimeString('zh-Hant', {hour12: false, hour: '2-digit', minute:'2-digit'}) ]] ~ [[ new Date( '2023-07-29 03:10:00+00:00' ).toLocaleTimeString('zh-Hant', {hour12: false, hour: '2-digit', minute:'2-digit'}) ]] zh-tw
加入行事曆 加入關注 加入關注 已關注

Tensor Operator Set Architecture (TOSA) 提供了一套常用於深度神經網路的全張量操作。其目的是使得在不同處理器 (CPU, GPU, NPU) 上運行的各種實現規範，能夠在 TOSA 層面上保持一致的結果。

講者

Odin Shen

Odin Shen

Business Development | AI edge computing specialist | SoC arch. exploration | Semiconductor functional safety expert | Cloud Native Computing | Co-founder TinyML Taipei | Embedded Maker | Kid's Coding Volunteer

Open Edge AI & TinyML 7SVMUC general (30mins)

ONNC on TinyML - Enhance MLPerf Tiny scores with ONNC

TR 615 [[ new Date( '2023-07-29 03:20:00+00:00' ).toLocaleDateString('ja', {year: 'numeric', month: '2-digit', day: '2-digit'}) ]] [[ new Date( '2023-07-29 03:20:00+00:00' ).toLocaleTimeString('zh-Hant', {hour12: false, hour: '2-digit', minute:'2-digit'}) ]] ~ [[ new Date( '2023-07-29 03:50:00+00:00' ).toLocaleTimeString('zh-Hant', {hour12: false, hour: '2-digit', minute:'2-digit'}) ]] zh-tw
加入行事曆 加入關注 加入關注 已關注

隨著機器學習這幾年在各領域的蓬勃發展，AI 開發者們也逐漸想在 MCU 上運行 AI Model ，而 TinyML 社群因此孕育而生，匯集了 MCU/MPU 硬體、系統軟體、模型開發者、應用開發者來研討如何強化 ML 在 Tiny Devices 的各式機會。另一方面， AI Benchmark 的組織 MLCommons 也注意到 AI 開發者們在 Tiny Devices 上的需求，而開發了 MLPerf Tiny ，其為目前用來測量 AI on MCU 上最權威的 benchmark，AI developer在挑選 MCU 時，首重的就是 MLPerf Tiny 跑分的成績。Skymizer 在今年六月公佈的 MLPerf Tiny 上 ONNC 的跑分成績在所有的 benchmarks上，效能與記憶體使用量均優於 MicroTVM。我們使用了兩種不同的 MCU – STM 與新唐 – 搭上兩種不同的作業系統 Zephyr 與 mbedOS，在效能上達到 6%~33% 不等的優勢。

在本篇演講裡，我們會先介紹 TinyML 社群以及市場趨勢，再來介紹 Skymizer 參與 MLPerf Tiny 的經驗、與 ONNC 如何以編譯器的角度來優化 AI Model 在硬體上運行的效能。

講者

Luba Tang

Luba Tang

Luba Tang is the founder and CEO of Skymizer Taiwan Inc., which is in the business of providing system software to IC design teams. Skymizer’s system software solutions enable AI-on-Chip design houses to automate AI application development, improve system performance, and optimize inference accuracy. Luba Tang’s research interests include electronic system level (ESL) design, system software, and neural networks. He had focused on iterative compilers, ahead-of-time compilers, link-time optimization, neural network compilation, and neural network optimization. His most recent work focuses on exploiting various types of parallelism from different accelerators in a hyper-scale system-on-chip.

講者

謝政道

謝政道

Dr. Cheng-Tao Hsieh is the Compiler Team Lead at Skymizer, leading the development of ONNC. His research expertise includes static analysis of peak power and automatic addition of redundant circuitry to tolerate delay variation, as well as experience in low-power high-level synthesis targeting FPGAs. Prior to joining Skymizer, he held a position in EDA at Qualcomm and earned a Ph.D. in Computer Science from NTHU. His professional knowledge and leadership contribute significantly to the team’s growth and product innovation.

講者

Peter Chang

Peter Chang

Peter Chang is the business development manager and the co-founder of Skymizer Taiwan Inc. His research interests span areas in operating systems, virtualization, and computer architecture. Currently, he focuses on topics in hardware/software co-design and benchmarking on Machine Learning. He is also devoted to participating in the MLPerf Tiny and TinyML communities. He was also the maintainer of SkyPat, an open-source performance unit-test suite, and ARMvisor, one of the Kernel-based Virtual Machine solutions on ARM architecture.

Open Edge AI & TinyML KZZWTS general (30mins)

Large Language Model Optimization with Intel OpenVINO Toolkit & Neural Network Compression Framework (NNCF)

TR 615 [[ new Date( '2023-07-29 06:00:00+00:00' ).toLocaleDateString('ja', {year: 'numeric', month: '2-digit', day: '2-digit'}) ]] [[ new Date( '2023-07-29 06:00:00+00:00' ).toLocaleTimeString('zh-Hant', {hour12: false, hour: '2-digit', minute:'2-digit'}) ]] ~ [[ new Date( '2023-07-29 06:30:00+00:00' ).toLocaleTimeString('zh-Hant', {hour12: false, hour: '2-digit', minute:'2-digit'}) ]] zh-tw
加入行事曆 加入關注 加入關注 已關注

生成式人工智慧 (Generative AI) 鋪天蓋地的進入我們的生活當中,裡面最關鍵的就是進行自然語言處理 (Natural Language Processing) 的大型語言模型 (Large Language Model/LLM)了! LLM動輒數億到上千億的參數量,對執行模型推論的設備所需投入的資金以及消耗的能源非常的可觀,也不是一般大眾可以負擔的起的. 把模型最佳化到可以運行在大家都可以四處取得的設備,尤其是Edge AI裝置,是AI民主化關鍵的推力. Intel OpenVINO toolkit 以及Neural Network Compression Framework (NNCF), 不只提供了許多模型最佳化的演算法, 例如Quantization, Pruning, …等，也提供了讓這些最佳化後模型得以用最高效率執行的軟硬體環境．這一節我們將透過一些實例的分享，讓大家瞭解模型最佳化的威力,以及讓LLM執行在Edge AI設備的方法.

講者

Chungyeh Wang

Chungyeh Wang

https://www.linkedin.com/in/chungyeh-wang-5bb23447/ Chungyeh Wang runs AI software & OpenVINO on Intel Deep Learning Accelerators customer enabling for Intel. He loves to dig into customer problems and solve with technology. I have been working on optimizing deep learning applications and collaborating with ecosystem to drive AI revolution.

Open Edge AI & TinyML 3QSNJ9 general (30mins)

手把手帶領多款國產Smart AI CAM與語音手勢辨識開發板

TR 615 [[ new Date( '2023-07-29 06:40:00+00:00' ).toLocaleDateString('ja', {year: 'numeric', month: '2-digit', day: '2-digit'}) ]] [[ new Date( '2023-07-29 06:40:00+00:00' ).toLocaleTimeString('zh-Hant', {hour12: false, hour: '2-digit', minute:'2-digit'}) ]] ~ [[ new Date( '2023-07-29 07:10:00+00:00' ).toLocaleTimeString('zh-Hant', {hour12: false, hour: '2-digit', minute:'2-digit'}) ]] zh-tw
加入行事曆 加入關注 加入關注 已關注

由國內瑞昱大廠與資策會極力推廣，比ESP32 CAM具有AI功能的升級開發板HUB8735(AmebaPro2)，增加內置NPU AI 運算引擎，手把手簡易修改程式即可操作屬於自己的AI模型，其還帶有802.11 a/b/g/n 雙頻Wi-Fi與BLE低耗電藍牙。藉由該議程帶領您了解另外還有奇景光電、新唐所推的SMART-AI-CAM，亦可用於語音、手勢等應用之no-code AI訓練國產開發板，當您遇到資安國安或是標案中指定使用國產晶片，何不來試試可免費申請試用的資策會國產開發板，如效果不錯政府還會幫您免費產品化。

講者

章育銘

章育銘

資策會講師與工廠智慧化接案人員，通訊大賽、中華電信等多項競賽得獎者，門薩成員。擅長自動化整合、AIoT、深度學習...，參與資策會合作多款國產開發板。

Open Edge AI & TinyML JXS9SZ general (30mins)

TinyML新玩法─揮揮手馬上就能變身簡報播放遙控器

TR 615 [[ new Date( '2023-07-29 07:20:00+00:00' ).toLocaleDateString('ja', {year: 'numeric', month: '2-digit', day: '2-digit'}) ]] [[ new Date( '2023-07-29 07:20:00+00:00' ).toLocaleTimeString('zh-Hant', {hour12: false, hour: '2-digit', minute:'2-digit'}) ]] ~ [[ new Date( '2023-07-29 07:50:00+00:00' ).toLocaleTimeString('zh-Hant', {hour12: false, hour: '2-digit', minute:'2-digit'}) ]] zh-tw
加入行事曆 加入關注 加入關注 已關注

一般播放PowerPoint簡報時都需要一組無線遙控器來控制播放下一頁動作，這裡使用運動手勢就能代替簡報遙控器。主要使用Seeed Xiao nRF52840 Sense(Arduino Nano33 BLE Sense同級產品）作成類似運動手環裝置，並利用開發板上的運動感測器及Edge Impulse TinyML雲端開發平台來訓練運動手勢模型，最後產生Arduino源碼，手動整合BLE HID功能，就能以手勢變成電腦上的按鍵按下，藉此來控制簡報播放。

講者

Jack OmniXRI

Jack OmniXRI

來自歐尼克斯實境互動工作室的神奇傑克，在TinyML領域另有一個小名「史蒂芬周」。

Open Edge AI & TinyML QF9FYU general (30mins)