Google Clips: Wearable Camera Able to Use Machine Learning to Capture Spontaneous Moments in Life
While impressed by Clips’ size and its underlying technology, I reserve judgement on the merits of the product and its utilitarian value in everyday life. Let us start by the engineering feat.
The wearable camera comes with a 12 MPixel image sensor protected by Gorilla glass. It is able to do autofocus adjustments, 130 degree field of view and captures at 15 fps. The Clips comes with 16 GB storage, supports many video formats as well as a myriad of interfaces (USB-C, Wi-Fi Direct, and Bluetooth LE). What sets this product apart from other wearable cameras is Its deep object identification capability (powered by machine learning). The camera uses CNN-based algorithms to recognize great expressions, lighting, framing, and faces (collectively referred to as “great moments” by Google). Through continuous learning process Clips’ ability to detect and capture such moments improves over time, in a way the device over time gets better in identifying and capturing the good stuff and refuses to waste storage on storing the mundane.
The computational beast responsible for the deep learning is an Intel Movidius Myriad 2 vision processor (VPU). Thanks to its low power dissipation the Clip can support an “always on” mode of operation. I am assuming the unit only performs inference locally and the actual training is done in the cloud. I personally don’t see much use for this gadget in my life but my taste has proven to be a contra indicator in the past. One notable exception is the adoption of Clips by parents with infants and small children. Putting personal preferences aside, I think this is a neat product and I can think of numerous industrial and scientific applications that can benefit from it.
ShopBot, . . . Hmmm?
While scanning some of the ways that eCommerce companies are utilizing cloud-based Nvidia GPUs, I was blown away by the capabilities of an app developed by eBay called ShopBot. The app is currently available on Beta but can be used via Facebook Messenger. The Android and IOS app will be coming soon. The support will also be extended Alexa, Google Home, and SIRI. As the name suggests, the app is intended to help customers to easily find and purchase the right products. The user has many ways of searching for their desired product and can place orders without leaving the messenger. Users can search by typing or saying a contextual search clause (e.g. find me a used SLR camera in fair condition under $1000 and leave out brand XYZ). Once a list has been identified, the buyer can fine tune the selection using verbal or textual feedback. Alternatively, the user can upload the image of a desired product and the app will quickly find an identical or similar product. Another super cool feature is the support of “Curated Shopping”. This is basically a fancy phrase describing a virtual unbiased salesperson that has indepth knowledge of multiple brands and models of a given product and can quickly point out the pros and cons of each selection.
Hard to imagine that all this can be done by GPUs running on a server cluster in a remote datacenter.
Banks are on the Forefront of Squeezing Value from Deep Learning
Banks are on the forefront of using Deep Learning to improve customer service and gaining operational efficiencies. Financial service companies have been making huge investments in big data and machine learning for years and are starting to enjoy sizable dividends from such investments. It all started from chatbot able to answer simple customer questions, but the scope has expanded massively in the past couple of years. Payment fraud detection, detecting cyber-attacks, malware, and money laundering are clearly on top of the list but it does not stop there. Natural Language Processing is used to learn from customer chat logs and utilize the findings to improve future transactions not to mention generating customer-specific wealth management advice. Gaining workflow efficiencies in banks is a new front impacted by AI. One can name loan application processing, automated ledger reconciliation, facilitating regulatory compliance are few examples.