Tech Shift Drives Investment — Deep Learning

In my last post, I discussed how the tech shift in hybrid cloud has shaped our thinking in startup investment. In this post, let’s take a look at the second tech shift — deep learning.

Deep learning neural networks have made significant progress in recent years, especially in the area of Natural Language Processing (NLP) and Computer Vision (CV). The improvement has accelerated in recent years such that the accuracy on typical tasks such as speech/object/facial recognition have reached and may soon exceed human performance.

While NLP and CV touch many direct applications, we believe the rapid advancement of deep learning will propel the following four significant investment areas:

Service robotics covers industrial/business robots whose primary tasks involve interaction with humans. IDC report shows worldwide robotics spend will be $103B this year with year over year growth of 25% over the next 4 years. Industry robotics dominates it with 70%. However the advancement in deep learning is enabling robots to work much more successfully in human-intensive environment. For example, Security robots(roving, flying or semi-stationary) that can drastically improve the cost effectiveness of human security patrol. We have made investment in one such company in Turing Video. There could be more use cases in delivery, retail/hospitality and office services.

Consumer robotics covers primarily in-home robots. We think that in-home robots will get a big boost due to the improvement of speech, vision and the significant cost lowering of a particular area of CV called SLAM. In-home robots are ready to take over more types of chores, securing the perimeter, monitoring the young and infirm, and start to manage the other less “smart” appliances. PerceptIn Robotics is one of the startups in this area invested by our prior fund.

Autonomous Driving can be considered a special kind of service robot. Car/truck/delivery vehicle driving is one of the most complex AI applications that combine more than a dozen different technology areas from control, sensor fusion, vision/perception, positioning, mapping, navigation, planning, simulation and etc. Many of these areas are starting to benefit from deep learning and become more versatile and human like. There are currently 56 companies registered with California DMV to conduct autonomous road test. We invested in JingChi as we believe they are the most likely to succeed in starting a robo taxi service in China.

Improved UI (User Interface) is an often overlooked area that will be profoundly impacted by deep learning. AI is not only the new UI, the very definition of user interface will be predicated on the level of intelligence. The world of computation has gone through several iterations of center(mainframe, cloud), edge(PC, smartphone) and networking improvements. Now UI has become the bottleneck. Even the well known O(n) computation complexity model has shifted from measuring against computation resources to that of human complexity. We believe deep learning based natural language/speech interface will increase its prevalence in enterprise customer services (we have invested in Percept.AI), and in enterprise robotic process automation (we have invested in jane.ai). In addition, ASIC chips designed to accelerate deep learning (we invested in IntEngine in this area) will enable more smart appliances with language and vision based user interfaces.

Stay tuned for the next topic…