Domainknowhow driven ADAS AR VR People Application driven

  • Slides: 14
Download presentation

未來物聯網多元應用場域與新興終端載具 將帶動軟硬整合系統的發展 Domain-knowhow driven ADAS AR & VR People 社交、健康… Application driven Software-led Adaptive

未來物聯網多元應用場域與新興終端載具 將帶動軟硬整合系統的發展 Domain-knowhow driven ADAS AR & VR People 社交、健康… Application driven Software-led Adaptive Cruise Control Smart Transportation 自駕車、車聯網、智慧 運輸/交通… Autonomous car Collaborative Robotic Servant Smart Home 智慧娛樂、節能… Building/Community 智慧建築、社區 Business/Manufacturer 智慧製造、物流… System-led Self-driving Bus Smart Lighting NB-Io. T Smart Meter Smart City 永續城市、智慧安防、 城市治理… Drone in Logistics Personal Drone Smart Garment 資料來源: 研院IEK(2017/01) Smart Devices Personal Mobility 行動寬頻、智慧旅遊、遠端控制… Copyright 2018 All Rights Reserved Surveillance Camera 2

傳統Io. T分析 具 vs. AIo. T Analytics Backwardlooking Hindsight with more Humancentric 資料來源:Gartner ;

傳統Io. T分析 具 vs. AIo. T Analytics Backwardlooking Hindsight with more Humancentric 資料來源:Gartner ; Information Builders ; Aria. Networks(2017) Forwardlooking Insights with more Machine. Centric Copyright 2018 All Rights Reserved 4

Why AIo. T analytics at edge/things? ? Balance between performance, price, and power consumption

Why AIo. T analytics at edge/things? ? Balance between performance, price, and power consumption Privacy 資料來源:Gartner(2017/11) Copyright 2018 All Rights Reserved 5

Deep. Mind reinforcement learning with Io. T reduces Google data center cooling bill by

Deep. Mind reinforcement learning with Io. T reduces Google data center cooling bill by 40% • Leveraging thousands of sensors within the data center, collecting data such as temperatures, power, pump speeds etc. , Google succeeded in reducing its cooling bill by 40%, and improving energy usage around 3. 5 times the computing power out of the same amount of power consumption. 資料來源:Gartner ; Deep. Mind(2016) Copyright 2018 All Rights Reserved 7

將AI導入Io. T三部曲 No need for tags and labels While dealing with insufficient pool of

將AI導入Io. T三部曲 No need for tags and labels While dealing with insufficient pool of labeled training data 資料來源:Gartner(2017/06) Copyright 2018 All Rights Reserved Reinforcement learning’s objective is to acquire a mapping between situations and actions. 9

AI領域發展方向與臺灣機會 ü AIo. T晶片市場預計到 2025年將達 390億美元,CAGR為 20%。 ü AIo. T少量多樣碎片市場特性提供臺灣新創參與全球End-to-End研發生態系並價值創造利基。 Ease-of. Use Scalability Time-to.

AI領域發展方向與臺灣機會 ü AIo. T晶片市場預計到 2025年將達 390億美元,CAGR為 20%。 ü AIo. T少量多樣碎片市場特性提供臺灣新創參與全球End-to-End研發生態系並價值創造利基。 Ease-of. Use Scalability Time-to. Market Audio/NLP Training Platform Noise Cancellation Visual/Video High Throughput Low Power & Low Footprint Predictive Analysis Behaviors Analysis ASIC (Cloud/Edge) Prescriptive Analysis Data Acquisition Sentiment Analysis Machine Comprehension 3 D Search Face/ Gesture Features Modeling AIo. T App/Service Real Time Incremental Learning Ultra low power & energy harvest Collaborative Learning In-Memory Compute Io. T Security 產業議題與展望: • AI的全面導入與應用將大幅提升企業對未來預測掌握之精準性以及營運決策的即時性,同時 提供客戶端更為貼切且適時適地適性的客製化服務。 • 根據Goldman Sachs的預測,2025年AI將為全球農業創造 200億美元的市場,並分別為零售 業以及能源業各節省540億以及140億美元的成本。 資料來源: 研院IEK(2018/06) Copyright 2018 All Rights Reserved 10

國際AIo. T新創方興未艾 AIo. T Start-ups創新焦點: Ultra low power & energy harvest In. Memory Compute

國際AIo. T新創方興未艾 AIo. T Start-ups創新焦點: Ultra low power & energy harvest In. Memory Compute Incremental Learning 資料來源: 研院IEK(2018/06) Collaborative Mobility Learning Io. T Security • Cybereason係由以色列軍方資安專家於 2012年所成立的新創公司,2018獲選CB Insights AI 100;拜Soft. Bank投資1億美 元之賜,目前已是獨角獸之一員。 • SHIELD AI 2015年成立於聖地牙哥,與 美國國防部以及國土安全部開發監控用 無人機,獲選為CB Insights 2018 Game Changer。 • Fog. Horn成立於 2014年,投資者包括 Intel、GE、Dell、Honeywell、Bosch等 大廠,2017並獲Gartner評選為Cool Vendor (Io. T Edge)。 • 英國AI晶片新創公司Graphcore著手研發 具記憶功能的運算元件,如同記憶與運 算在大腦的同一區間執行一般(記憶功能 存在於神經元/突觸之中);Graphcore亦 入選CB Insights AI 100以及Game Changers 2018。 Copyright 2018 All Rights Reserved 11