You just have to submit the description of your problem, some pointers as to how to get started, and provide lightweight supervision along the way (occasionally answer questions, provide feedback, suggest experiments to try...). What you get out of this: - Innovative solutions to research problems that matter to you. - Full credits for the value you provide along the research process. We are looking for both deep learning research problems, and problems from other fields that could be solved using deep learning. Note that the information you submit here may be made public (except for your contact information). We will create a website listing the problems submitted, where people will be able to self-organize into teams dedicated to specific problems.

But deciding to build an app in the first place was driven by a broader purpose. As one doctor put it, "If you want to do world-class sociology research, build Facebook."

We @ProjectJunoAI are big fans of landscapes. That's why we've created a machine intelligence landscape focused entirely on Europe [1]. Europe deserves a landscape of its own to highlight its talent and expertise. Until recently, its contribution to the innovation and commercialisation of machine intelligence technologies has been under-appreciated. We now see growing self-confidence borne of the success, and continued presence, of local acquired startups like VocalIQ, Swiftkey, Deepmind, Magic Pony Technology, and PredictionIO.

Google's DeepMind announced the WaveNet project, a fully convolutional, probabilistic and autoregressive deep neural network. It synthesizes new speech and music from audio and sounds more natural than the best existing Text-To-Speech (TTS) systems, according to DeepMind. Speech synthesis is largely based on concatenative TTS, where a database of short speech fragments are recorded from a single speaker and recombined to form speech. This approach isn't flexible and can't be adjusted to new voice inputs easily, often resulting in the need to completely rebuild a dataset when there's a desire to drastically alter existing voice properties. DeepMind notes that while previous models typically hinge around a large audio dataset from a single input source, or single person, WaveNet retains its models as sets of parameters that can be modified based on new input to an existing model.

The artificial intelligence offshoot of Google has paired up with University College London Hospitals NHS Foundation Trust for a research project. DeepMind Health announced that it will be receiving anonymised data from the trust for a research partnership into head and neck cancer. The five-year collaboration will use around 700 anonymised CT and MRI scans of former patients, dating back to 2008, and additional data relating to approximate age, anatomy location, cancer type and radiotherapy received. Currently before radiotherapy can be administered, clinicians take up to four hours to identify and differentiate between cancerous and healthy tissues on CT and MRI scans of head and neck cancer patients. This process is called segmentation.

Neural networks are very powerful tools to classify data but they are very hard to debug. Indeed, they do a lot of computation with low level operations so they are like black boxes: we provide inputs and get outputs without any understanding on how the neural network is finding the results. Few years ago some scientists found ways to delve into the networks used for image categorization. Instead of doing backpropagation on weights like during the learning phase of a neural network, they did backpropagation on the images themselves: in the example below (edited from CS231n), considering x are inputs and w are weights, each learning step, the gradient (red) is applied to the x instead of the w. In this article, we will use the method and code from Google, Simonyan, Yosinski and Chollet to try to visualize the classes and convolutional layers learnt by popular neural networks. The code provided in this article uses the Keras library.

Startup Drive.ai is revealing its product and strategy for the first time, and the autonomous driving tech company is looking not only to create the best hardware and software to enable self-driving cars, but also to make sure those cars communicate with people outside of the car in the most effective way possible. Core to Drive.ai's approach is using deep learning across the board in its autonomous driving system, which means they're teaching their self-driving cars somewhat like how you'd teach a human. That involves providing a host of examples of situations, objects and scenarios and then letting the system extrapolate how the rules it learns there might apply to novel or unexpected experiences. It still means logging a huge number of driving hours to provide the system with basic information, but Carol Reiley, co-founder and president of Drive.ai, "We are using deep learning for more of an end-to-end approach. We're using it not just for object detection, but for making decisions, and for really asking the question'Is this safe or not given this sensor input' on the road,'" Reiley explained.

Have you seen the work of Friedlander and Schmidt, and my follow up paper (shameless plug, toot toot)? Though our analysis is restricted to convex functions, there is also a notion of "sharpness" of minima which is appears as the condition number of the problem.

AI has achieved a win experts once thought wasn't possible. Harvard University was recently awarded a 28 million grant to discover why human brains are so much better at learning and pattern recognition than artificial intelligence (AI). Dispensed by the Intelligence Advanced Research Projects Activity (IARPA), the funding will fuel a quest to make AI systems faster, smarter, and match or outperform human neural networks. The steep challenge in this quest is the enormous complexity of the human brain and its billions of neurons and trillions of synaptic interconnections with electrochemical signaling. The other challenge: There is no accepted theory of mind that describes what thought--the gist of intelligence--actually is.

Machine intelligence has been the subject of both exuberance and skepticism for decades. Beginning in the 1950s, Marvin Minksy, John McCarthy and other key pioneers in the field set the stage for today's breakthroughs in theory, as well as practice. Peeking behind the equations and code that animate these peculiar machines, we find ourselves facing questions about the very nature of thought and knowledge. The mathematical and technical virtuosity of achievements in this field evoke the qualities that make us human: Everything from intuition and attention to planning and memory. As progress in the field accelerates, such questions only gain urgency.