NJ35_web

37/40

［フォーラム後記｜Forum Epilogue］自動運転などディープラーニング関連のマーケティングの第一人者であり、GPUやAI全般に造詣が深いNVIDIAの佐々木氏に、これまでとこれからのお話を分かりやすく説いていただきました。　昨今、パソコンでもそれなりのGPUを実装することで、ディープラーニングが我々エンドユーザーにも比較的手軽になってきました。昔のスーパーコンピュータがスマホになるなど、ソフト面でも大きなダウンサイジングがすでに起きています。　2020年オリンピックに向けて自動運転の実用化が進められていますが、自動車が自律車に変貌するなかで社会がどう変わっていくのか、AIによる自律制御は都市や建築にどんな影響を与えるのか、今回の講演で垣間見れたような気がします。身の回りには、車のほか、エレベータや掃除機、警備ロボット、ドローン、さらにはスマートスピーカーなど自分で判断し適切に実行するものがあふれていきます。そのなかで、スマートビルディングやスマートシティが、人間を包み込み、リアルとバーチャルの知能的統合により豊かな生活を実現するものになると確信しました。　テクノロジーはすでに我々の傍にあり、あとは安全・安心を確保する法整備が追い付くよう、業界が横串で協業することが重要であると考えます。 ［ファシリテーター：吉田 哲］An expert marketer of deep learning and AI for such purposes as autonomous driving and well-versed in GPU and AI technology in general, Mr. Sasaki explained to us the past and future of this technology in an accessible manner. Having GPUs installed on personal computers makes deep learning more accessible to end-users. Downsizing of apps has already begun̶the super computers of yesterday are now in our hands as smartphones. Practical application of autonomous cars is underway toward realization by the time of the 2020 Olympics. In this lecture, we caught a glimpse of how society might change when automobiles become autonomously driven cars and of the effects AI-driven autonomous cars could have on cities and architecture. In addition to cars, our lives will fill up with objects that make their own decisions and act accordingly̶elevators, vacuum cleaners, security robots, drones, and smart speakers. The lecture made me confident that humans will be surrounded by smart buildings and smart cities, and better and richer lifestyles will result from the integration of real and virtual intelligence. Technology is already by our side. What’s important now, I believe, is for the business world to cooperate across various fields to enable laws and regulations to keep pace with technology. [Facilitator: Tetsu Yoshida]Expanding the Range of Artificial IntelligenceDeep learning is getting attention as an effective method of machine learning to bring AI to reality. Graphic process-ing units provide the hardware that support deep learning. Originally developed for image processing, these units are the devices that enable the creation of virtual reality in the graph-ics world. The combination of GPU computing capacity and AI/deep learning research has made them widely applicable in various fields. Let me list a few examples. First, in the promising field of medicine where AI appli-cations are attracting much interest, development of AI is underway that can accurately predict the onset of 80 different diseases. In the field of meteorology, after learning from an enormous number of satellite images, AI can now deter-mine the effects of greenhouse warming and CO2 with a high degree of precision based on the images. In the world of cyber security, AI is used in antivirus software that is able to detect malware subspecies. At call centers, AI converts conversation recordings into text, determines the subject product of the call and also determines the emotion of the caller using voice recognition. Highly versatile, AI is able to make determinations appropriate for each particular situation as it learns. As more data is added, precision is improved by training the model to match answers given by humans. Edge Computing and Autonomous VehiclesWhile cloud computing shares computing functions via the Internet, edge computing uses small computers mounted on individual devices. We have developed a GPU-mounted com-puter the size of a credit card. Frequently used on drones and on spectacle-shaped devices that help visually impaired per-sons travel using image recognition, small GPUs have poten-tial for changing people’s everyday lives for the better. Edge computing is needed for self-driving vehicles. We have developed three kinds of applications for self-driving cars that utilize deep learning. “Drive AV” combines onboard image recognition and map information for precise autono-mous driving while avoiding danger. “Drive IX” uses image recognition to detect eye movements and driver condition, and prevents the driver from falling asleep at the wheel to increase driving safety and comfort. “Drive AR” uses aug-mented reality to point out dangerous elements such as bicycles appearing in rear-view mirrors, etc. Small comput-ers support autonomous driving on many levels from driver assistance to unmanned operation. I believe the possibilities for AI made possible by deep learning can only expand in various fields from now on. 左図：DRIVE AR 搭載イメージ。ディープラーニングの認識結果を、運転に役立つ情報としてメーターパネルディスプレイやバックミラーに反映。（提供：NVIDIA）Le: Images displayed in the Drive AR shows deep-learning recognition results useful for driving information on the instrument panel display and in back mirrors (Courtesy of NVIDIA).372018 SUMMER35FORUM