How will AI techniques be applied to post-covid 19 world?
Working together with cloud-based AI, on-device edge AI ensures more privacy, increases computing speed with low latency, and lowers total costs to integrate AI into everyday devices.

人工智慧的技術正在加速發展中，如何利用神經網路處理大量的資料，並且在邊緣或終端裝置做出即時推論，成為近年來熱門話題。也正是如此，記憶體變成關鍵且不可或缺的角色。華邦所提供的GP-Boost DRAM系列產品有著低功耗高頻寬的特性，正符合邊緣或是終端運算所需要的規格，並帶來相當高的效益。

The cloud AI with high performance training & Inference is changing the world. From sky to the ground, the AI everywhere can be realized by the edge/endpoint AI device in 5G infrastructure. However, the lots of IoT device and autonomous vehicle bring tremendous data. How to manage these data though AI Neural Network? The memory play the key role no matter NN data heavy traffic even near memory computing. Hence, Winbond offer GP-Boost DRAM solution for low power and high bandwidth feature. We believe GP-Boost DRAM can enhance edge/endpoint AI to be real world in coming future.

Arm 致力於在每個應用領域，例如影像、預測文字、以及語音等，提供人工智慧平台，今天的演講將會分享 Arm最新的硬體、軟體解決方案，以及與生態圈夥伴一起合作的成果。

Arm technologies enable many computing platforms around the world to benefit from machine learning (ML) features, such as video, predictive text, and speech. This session will present the latest Arm ML solutions, from hardware IP, software framework, and ecosystem collaboration.

為了因應前所未有且不斷變化的網絡安全挑戰，愈來愈多的企業以人工智能來改善安全防護並減少被攻擊行為。 但是，由於與其他技術相比，人工智能需要更複雜的數據及資料收集，因此在導入時更需要小心而謹慎。這也使得保護演算法和敏感數據及資料成為AI技術使用的關鍵，反之，這樣的技術會成為資安的破口，而不是保護。
從資訊安全的整體生態來說，重要的是建立一個全面的物聯網安全生態系統。而該生態系統由五個主要結構組成：安全法規和安全認證，雲安全性，平台安全性，軟體安全性和硬體安全性。 每個結構層級除了都需要確保有效的安全保護之外，大眾對於使用符合安全標準產品的意識，更是必不可少的。

In response to the dynamic and unprecedented challenge of cybersecurity, Artificial Intelligence has emerged to improve security posture and reduce breach. However, the use of AI based tools for cybersecurity should be more cautious as it requires more complex data than other technologies, which makes protecting algorithms and sensitive data a key.
From the perspective of security ecosystem, it is important to establish a comprehensive ecosystem for IoT security. The ecosystem is built up with security regulations, security certification, cloud security, platform security, software security and hardware security. Each layer needs to be integrated to ensure effective security protection. Furthermore, the security awareness is rather essential for the public to use security-qualified products.

儘管從數據中心到智能揚聲器，同時包含診斷和智能學習機器在內的人工智慧 (AI) 已經應用於多種類型的設備中，但由於有限的能源，記憶體，及伴隨周邊所需的其他資源。即便近期AI芯片的發展突飛猛進，然而終端產品在閒置狀態下仍消耗著比例過高的電量。現今最佳的使用場景是芯片本身消耗系統整體所需的功率再來執行“後台”的任務已期對後段執行造成零負擔。終端以外靠電池發電的產品若能夠進行推理和模仿人類智能，成為實現外來物聯網最重要也是最大的挑戰。Ambiq憑藉其獨特的SPOT平台解決了這些問題，從而使芯片上的完整系統能夠在超低功耗預算內進行傳感，數據存儲，分析，診斷和通信。

Although Artificial Intelligence (AI) , including inferencing and machine learning, has already been in many types of equipment, from data centers to smart speakers, virtually all AI workloads are still running in the Cloud or selected Edge devices because of the energy, memory, and other resources required to support them. Even with the recent explosion in AI chips' development, today's AI chips still consume far too much power for Endpoint use cases even while sitting idle. Today, the best available chips will consume the entire system power budget to tend "background" tasks, leaving no power budget for inference. Enabling battery-powered Endpoints beyond the Edge to run inference and mimic human intelligence becomes the most important and the biggest challenge to realizing the IoT future where everything is connected. Ambiq has addressed these problems with its unique SPOT platform, allowing complete systems on a chip capable of sensing, data storage, analysis, inference, and communications, all within an ultra-low power budget.

面對運算邊緣化的智能時代，恩智浦在硬體上除了MCU，AP還有軟體上的若干支持，讓開發AI相關領域應用上顯得更得心應手，此舉不僅縮短開發時間，也讓公司的開發依著恩智浦的各個平台，更系統化、智能化。

人工智慧快速地從雲端運算改變到終端裝置應用，未來在人臉/影像辨識、聲紋辨識等應用將大量使用到邊緣運算來加速整體運作或相關的不同終端設備佈署。針對各種不同品質、可靠度、封裝別等需求之快閃記憶體隨之衍生出來，而在快閃記憶體耕耘多年的華邦電子可以針對不同之應用提供最佳的解決方案。

The AI is highly and quickly developed and considered to be implemented in multi applications. Such like code/model storage, voice storage and image recognition in edge computing and endpoint applications. The reliability, quality requirement, high performance and various package types are required. Winbond could offer varies flash selections to fit different AI application field requirement.

Introduce Kneron's product and how data flow inside the chip and what is the logic of memory access correlation to the performance.
Furthermore, update Kneron's commercial application and also explain our next step of full stack edge AI.

講者以過往資安所研發專案成果與經驗為例，分享如何透過AI/機器學習建構並協助資安維運的日常業務，最後帶出AI/機器學習技術導入可能產生的誤導例(Adversarial AI)等威脅與應對建議。

The speaker took the results in the past research and development projects as an example, shared how to construct and assist the daily business of cybersecurity through AI/machine learning. Besides, the speaker also brought out the adversarial example threats and other risks that caused by the AI/machine learning technology. Finally, the countermeasures and suggestions would discuss with audiences.