可穿戴裝置應及時收集資料，切身考慮使用者需求，隻有這樣，産業才能實作真正的商業化發展。
近日，由工業和資訊化部人才交流中心主辦，江北新區IC智慧谷頂山産業園協辦的“芯動力”人才發展計劃國際名家講堂系列最新活動在南京浦口區舉辦。作為活動的全程支援媒體，鎂客網受邀對本次講座的專家導師瑞士洛桑聯邦理工學院副教授David Atienza博士進行了一次專訪。
左起：工信部IC Power産業合作組王喆、洛桑聯邦理工學院David Atienza博士、鎂客網主編王剛
以下是鎂客網對David Atienza博士的專訪實錄：
鎂客網：David 您好，我們了解到，您的研究主要是針對“高性能多處理器系統級晶片（MPSOC）和低功耗嵌入式系統的設計方法”的，包括針對2D和3D MPSoCs的新型熱感覺設計，以及針對無線人體傳感器網絡，動态記憶體管理和互連層次結構優化的設計方法和架構。可以請您先介紹一下您在可穿戴裝置的功耗和系統優化方面取得了哪些最新的進展嗎？
David：近期，我們已經在EPFL的嵌入式系統實驗室（ESL）中取得了非常重要的進展了。首先，我們證明了近門檻值多核心計算架構是更加節能的，它能使單核微控制器能夠通過處理下一代智能型可穿戴裝置的複雜生物信号，甚至可以在穿戴式系統中運作機器學習算法，而不用将所有資料發送到遠端雲計算系統中。此外，在ESL的系統層面，我們證明了壓縮感應是可應用于基于ECG監測下的可穿戴裝置的，它可以顯著的節省我們在檢測生物信号和采樣所需要耗費的精力。
鎂客網：對于可穿戴裝置的商業化前景，您認為目前哪些方向的應用已經成熟或具有可預見的成熟市場環境？
David：我認為，在過去幾年時間裡，面向健康和消費者市場的大多數現有可穿戴裝置的應用方向并不是很正确的，因為這些裝置缺乏對使用者的測量或使用者提供的回報資料及臨床驗證，這也是大多數不願接受這些裝置的原因之一。但我認為，這一領域的産品正在迅速朝健康、可靠、經認證的方向發展，在系統保障使用者隐私和健康的前提下，面向健康和消費者市場的可穿戴裝置是可具有成熟的市場的。
鎂客網：我們一直在提“可穿戴裝置”的概念，前幾年這一類型的産品也大紅大紫過，其中最為突出的就是“智能手環”。但是它們卻迅速沒落了，全球出貨量大幅下滑，生廠商也大批倒閉或重組，您認為出現這種局面的原因有哪些？這其中除了它的實用性不足之外，當下這些産品在技術方面有哪些是急需改進的？
David：其實，當下可穿戴裝置産業下滑的情況與我之前的預期是一緻，以“智能手環”這一領域為例，大多數的産品是缺乏臨床驗證的，這使得很多使用者對手環系統給出的資料持質疑态度，例如完成的步驟數量或每天消耗的熱量等，這也是此類産品銷售量不理想的重要原因之一。
不過，如果這些裝置商可以很好的認識到這一點并對他們的産品進行改進的話，那麼我認為這一領域還是很有發展潛力的。此外，使用者界面方面，研發商應該多加考慮産品的适用和便捷性，并盡力使産品具備手機不可替代的功能，才有可能無讓智能手環真正、穩定的商業化。
鎂客網：那麼，您認為可穿戴裝置是有必要作為單獨産品出現，還是作為日常用品的一個特性出現呢？
David：可穿戴裝置是物聯網中的一個關鍵組成部分，對于我來說，它們就是物聯網的基礎，是接物聯網和人類的第一個環節。是以，IoT裝置的首次內建、互動是開始于具有可靠、舒适性的可穿戴裝置的，進而才與其他IoT裝置進行互動，最終讓智能城市、智能建築、工業4.0等成為現實。
鎂客網：我們注意到，您在課題中特别提到了“特征提取，機器學習和資料記錄”，那麼您認為，目前可穿戴裝置對資料分析的準确性如何？距離實際應用還要多久？
David：我要強調的是，可穿戴裝置從概念的提出到現在取得可在運作時提取生物信号的複雜功能這一重大進展僅有幾年的時間，是以，這是一個新的領域。而讓可穿戴裝置“聰明”起來是取決于真正适用的使用者和應用場景的，例如，我們近期已經在ESL-EPFL中證明，在心血管監測中，将可能患有心律不齊疾病的病人的ECG（心電圖）輸入裝置後，其識别的正确率可達到95-96％。是以，我認為這一新的可穿戴裝置系列（包括機載機）正在快速準備就緒，以便在獲得FDA（食品藥品監督管理局）認證或CE标志（一種安全認證标志）後被普通使用者使用。可以樂觀地認為，這一領域的産品是可以在未來幾年内在市場上全面鋪開的。
另一方面，像腦電圖分析這樣的領域，其在開發嵌入式機器學習方面有着很多的難題，是以，隻有通過與雲計算服務相連接配接，才能在短期内讓可穿戴裝置可以在機器學習算法上運作起來，進而為可穿戴裝置提供正确的訓示。
鎂客網：在您的設想中，終極的可穿戴裝置應用場景，是怎樣的？
David：在我看來，終極的可穿戴裝置是真正“聰明”、可自主供電的，并且，它們應該是可适用于每一個人。此外，這些可穿戴裝置應該可以給予使用者足夠的安全感。
是以，我所預計的終極的可穿戴裝置在形态上可能就是一個簡單的小貼紙，隻要貼在皮膚上就可以直接與大腦溝通、交流，不需要我們輸入一個字。雖然這對于在當下的技術研發人員來說是極具挑戰的，但我相信在未來，這一定會成為現實。
1, Can you tell us something about the latest developments you have made in the power consumption and system optimization of wearable devices?
In my Embedded Systems Laboratory (ESL) at EPFL, we have made very important recent developments that prove that near-threshold multi-core computing architectures are more energy-efficient that single-core microcontrollers for complex bio-signal processing in next-generation smart wearable devices, and that they can be even used to run machine learning algorithms on-board in wearable systems instead of sending all the data to remote cloud computing systems (which requires a lot more energy than computing on-board actually). Also, at the system level at ESL we have been the first ones to demonstrate that compressive sensing can be used in ECG-based monitoring in wearables to dramatically reduce the energy spent in bio-signal sensing and sampling.
2, For the prospects of the commercialization of wearable device, do you think the current direction of the application have matured or have a predictable mature market environment?
I believe the application direction of most of current wearables in the last years (oriented towards wellness and consumer markets) is not the right application to make a difference, as these devices lack clinical validation on what they measure or provide as feedback to the user. Therefore, many people start being reluctant to accept them any longer, but I think this area of products is rapidly evolving towards truly health-reliable certified devices, and it will have a mature market environment in preventive (and personalized) healthcare if the systems are validated well and do not bother the user (i.e., we should not realize they are there).
3, We have been talking about the concept of "wearable equipment" and this kind of product also flourished a few years ago, one of the most popular is the "smart bracelet." But they are declining rapidly, the global shipments fell sharply and many manufacturers also closed down or went through reorganization. What makes this happen? Which point of this product is urgently needed to improve in addition to its lack of practical use?
This situation is very much aligned with my previous answer because “smart bracelet” are one clear example of wearables where the lack of validation has made many customers believe that the current systems are not reliable in what they indicate (e.g., amount of steps done or calories spent per day, etc.) so do not buy them anymore. Thus, this is one clear example where the real validation and adaptability of the devices, as these devices lack clinical validation on what they measure or provide as feedback to the user. Therefore, many people start being reluctant to use them any longer. However, I think this family of products will gain ground again if the devices are well tuned to be reliable. On the other hand, a key point that needs to be improved is the user interface, as it is one of the main limitations for a practical use (e.g., how are you going to write an e-mail or reply in an elaborated way to a message that somebody has sent you?). Until the smart bracelets (or smartwatches) are not able to remove the need for a phone completely, they will not reach a real stable commercial success.
4, Just continue, do you think that wearable devices should be used as a separate product, or as a feature of daily use?
I believe wearables should (and will) become a (normal) feature of our daily life, as the smartphone has done in the end. It is just a matter of time and finding the right features for them so that people do not see them as “funny gadgets” but rather elements that improve our quality of life.
5, What about the relationship between wearable devices and IoT?
Wearable devices are a key component in the idea of IoT because for me they are at the foundations of IoT. Indeed, they must create the first link between IoT and human beings by allowing the other connected things (not in contact with our bodies) to understand what we feel or need or want. So the first integration of IoT devices starts on having reliable (and comfortable) wearable devices that can interact with the other IoT devices to make smart cities, smart building or appealing Industry 4.0 environments (to mention a few) interacting with humans.
6, We note that you specifically mentioned the "feature extraction, machine learning and data records" in your subject, so I want to ask how accuracy the data analysis result of wearable device is and how long it can be practically adapted?
Note that it is already a major progress (only done in the couple of years) that wearables can run machine learning by extracting at run-time complex features from biological signals. So this is a new field to make wearable devices truly adaptive to the person wearing them (so “smart” overall) and the current accuracy that these devices can reach depends on the application. For instance, in cardiovascular monitoring, the possible ECG inputs can be identified as correct or having a possible arrhythmia in the order of 95-96%, as we have proved recently at ESL-EPFL. So this new family of wearable devices including on-board machine are getting ready fast to be use in the general audience (after passing the corresponding FDA or CE mark, depending on the market in US or Europe). So I am optimist that we can have them on the market in this area in the next couple of years. On the other hand, other areas like EEG analysis are having more difficulties to develop embedded machine learning, so wearables having this support will only be deployed in the short term (4-5 years in my view) by connecting them to cloud computing services, where these machine learning algorithms will be running on the background and providing the indications to the wearables to adapt to their results.
7, Can you describe how the ultimate wearable device application scene will be in your vision?
In my mind the ultimate wearable devices will need to be really smart, autonomous (do not need any battery) and able to adapt to each person. Moreover, they will not be disturbing us at all (so almost to the point of being almost “transparent”). Therefore, in my view they could (eventually) have the shape of a simple tiny sticker that we put on our skin and can interact with use by communicating directly with the brain without us pronouncing a word. I know this is really challenging (or essentially impossible) with our current technology, but I am confident that we will progress enough to make it happen in the future.
原文釋出時間：2017-04-27 18:40
本文作者：伶軒
本文來自雲栖社群合作夥伴鎂客網，了解相關資訊可以關注鎂客網。