1 – Bots May be Cool, but DigitalGenius Thinks it Found a Better Way
London- and New York-based DigitalGenuis released a new product this week, Human+AI Customer Service Platform, that links to customer service platforms like Zendesk and Salesforce and supplements the chatbot experience. The service analyzes customer service logs from emails, chats, and other social interactions in order to assess the most common issues and determine the best or most likely response by the automated system. Each response has a ‘confidence threshold’, or how likely the response is to be accurate. The customer then decides the threshold with which they’re most comfortable receiving an automated response; any responses below the threshold are routed to a human agent. In addition to the rollout of its new hybrid customer service model, DigitalGenuis also raised another $4.1 million in venture funds from companies like Salesforce Ventures, Bloomberg Beta, and Singularity Investments.
(Read the full article on TechCrunch)
2 – Artificial Intelligence Can Now Design Realistic Video and Game Imagery
U.K.-based startup Magic Pony Technology has developed a machine learning neural network that can sharpen pixellated images and generate realistic video game backgrounds and textures based on earlier examples. What’s unique about Magic Pony’s approach is that instead of using manually labeled examples, the AI technology recognizes statistical patterns in high- and low-resolution examples and self matches those with features like edges, straight lines, and other textures. Furthermore, the company was able to develop this technology using an ordinary graphics processor, something that hasn’t been done before and that could open up further applications for the technology in smart phones, gaming, and virtual reality.
(Read the full article on MIT Technology Review)
3 – Artificial Intelligence for Everyday Use: Coming Soon
Programmers at Reactive Inc. created a machine-learning based software able to decipher handwritten Japanese, one of the more challenging orthographies due to its intricate form; more impressive is the fact that all four engineers had almost no knowledge of the language. Reactive’s engineers are one of the most recent groups to showcase the increasing ease-of-access to machine learning by more researchers, scientists, and developers in the field. Deep-learning technology can now be accessed on cloud-based platforms offered by several technology companies, including Microsoft and Nvidia. Seishi Okamoto, a project director at Fujitsu Laboratories Ltd., stated,
“The fact that this technology can side-step domain expertise gives it a big advantage in speed and scalability when it comes to business applications.”
More use of deep learning by smaller companies in various sectors promises to speed up AI technology in the next 10 years, especially as investment in AI startups are slated to grow to $310 million in 2015.
(Read the full article on Bloomberg Technology)
4 – Microscope uses artificial intelligence to find cancer cells more efficiently
UCLA scientists at the California NanoSystems Institute have developed a new microscope and deep learning software for better identifying cancer cells in blood samples. The photonic time stretch microscope is capable of more quickly imaging cells, and deep learning software can find cancer cells with 95 percent accuracy. The study, published in Nature Scientific Reports, was led by Professor Barham Jalali, doctorate student Claire Lifan Chen, and postdoctoral fellow Ata Mahjoubfar. Photonic time stretch was developed by Jalali, and the microscope is just one of many potential applications for the technology. In the paper’s findings, the researchers state that the technology could allow for quicker and earlier detection of cancer, as well as facilitation of new treatments.
(Read the full article on UCLA Newsroom)
5 – Announcing TensorFlow 0.8 Adds Distributed Computing Support
On Wednesday, Google announced an updated model of TensorFlow, which now includes oft-requested distributed computing support, along with a package of resources that allow users to train distributed models on proprietary architectures. Distributed TensorFlow can be leveraged with the recent release of the Google Cloud Machine Learning platform, which allows for more powerful training and serving of TensorFlow models. Among other features, the new model also includes new libraries for defining novel distributed models. Google continues to research ways to improve its distributed training services, and will share any algorithmic and engineering improvements via the TensorFlow community on GitHub.
(Read the full press release on Google Research Blog)
Image credit: Reactive Inc.