Deep learning can be understood as modeling high-level abstractions using data and a set of algorithms, with a deep graph and multiple processing layers of linear and non-linear equations. While it is largely a mathematical tool, the things that deep neural networks can do have surprised mathematicians. How can networks arranged in layers be quick to perform human tasks like face and object recognition if it has to go through layers of computations? This has perplexed mathematicians for sometime. However, what mathematics cannot makes sense of, physics explains simply.

Ever-increasing workload combined with dramatic reimbursement changes are driving development of methods that help clinicians gain value from medical images more efficiently. The brightest hope is artificial intelligence, fueled by advances in deep learning / convolutional neural networks. While the big vendors have made significant investments (Watson Health, GE Healthcare, Siemens etc.) and the usual academic groups are well represented, there is considerable activity at the smaller end of the scale. In preparation for the SIIM Conference on Machine Intelligence in Medical Imaging event in Alexandria, 12th -13th September we pulled together a list of the various start-ups working on machine learning solutions for medical imaging. It was surprising to see how many more have come out of stealth mode since RSNA last year – how many more are waiting in the wings?

This paper presents an end-to-end framework for task-oriented dialog systems using a variant of Deep Recurrent Q-Networks (DRQN). The model is able to interface with a relational database and jointly learn policies for both language understanding and dialog strategy. Moreover, we propose a hybrid algorithm that combines the strength of reinforcement learning and supervised learning to achieve faster learning speed. We evaluated the proposed model on a 20 Question Game conversational game simulator. Results show that the proposed method outperforms the modular-based baseline and learns a distributed representation of the latent dialog state.

Recurrent neural networks are one of the staples of deep learning, allowing neural networks to work with sequences of data like text, audio and video. They can be used to boil a sequence down into a high-level understanding, to annotate sequences, and even to generate new sequences from scratch! The basic RNN design struggles with longer sequences, but a special variant – "long short-term memory" networks – can even work with these. Such models have been found to be very powerful, achieving remarkable results in many tasks including translation, voice recognition, and image captioning. As a result, recurrent neural networks have become very widespread in the last few years. As this has happened, we've seen a growing number of attempts to augment RNNs with new properties. Individually, these techniques are all potent extensions of RNNs, but the really striking thing is that they can be combined together, and seem to just be points in a broader space.

In this special guest feature, Rob Farber from TechEnablement writes that modernized code can deliver significant speedups on machine learning applications. Benchmarks, customer experiences, and the technical literature have shown that code modernization can greatly increase application performance on both Intel Xeon and Intel Xeon Phi processors. Colfax Research recently published a study showing that image tagging performance using the open source NeuralTalk2 software can be improved 28x on Intel Xeon processors and by over 55x on the latest Intel Xeon Phi processors (specifically an Intel Xeon Phi processor 7210). For the study, Colfax Research focused on modernizing the C-language Torch middleware while only one line was changed in the high-level Lua scripts. NeuralTalk2 uses machine learning algorithms to analyze real-life photographs of complex scenes and produce a correct textual description of the objects in the scene and relationships between them (e.g., "a cat is sitting on a couch", "woman is holding a cell phone in her hand", "a horse-drawn carriage is moving through a field", etc.) Captioned examples are show in the figure below.

With new neural network architectures popping up every now and then, it's hard to keep track of them all. Knowing all the abbreviations being thrown around (DCIGN, BiLSTM, DCGAN, anyone?) can be a bit overwhelming at first. So I decided to compose a cheat sheet containing many of those architectures. Most of these are neural networks, some are completely different beasts. Though all of these architectures are presented as novel and unique, when I drew the node structures… their underlying relations started to make more sense. One problem with drawing them as node maps: it doesn't really show how they're used. For example, variational autoencoders (VAE) may look just like autoencoders (AE), but the training process is actually quite different. The use-cases for trained networks differ even more, because VAEs are generators, where you insert noise to get a new sample. AEs, simply map whatever they get as input to the closest training sample they "remember".

Baidu Inc. has established an early stage investment unit named Baidu Venture to back artificial intelligence, virtual reality and augmented reality projects. The fund will initially seek to raise US 200 million in capital for the effort. Robin Li, chairman and CEO of Baidu, will lead the venture unit as chairman and participate in project evaluation and investment decisions, Chinese media reported. The new venture unit will be independently run from Baidu's existing investment teams, allowing it to make fast decisions without having to go through complicated internal approval process. Baidu established a deep learning research institute as early as 2013, making artificial intelligence a key strategic focuse to shape the company's future.

Marketing departments are failing implement the most effective strategies [#ChartoftheDay] - Smart...

Drones and robots are getting computer vision with higher-resolution cameras and artificial intelligence to recognize objects and images. Many are made with developer boards like Nvidia's Jetson TX1, which provides the smarts for auto-navigation and collision avoidance. TX1 has the horsepower to process live image feeds, and software tools to instantly analyze and provide context to visuals. The TX1 is now a lot faster and better equipped to handle AI and image processing. Nvidia's new Jetpack 2.3 software tools for TX1, announced on Tuesday, are a major update that doubles the deep-learning performance of the board.

Microsoft researchers have reached a milestone in the quest for computers to understand speech as well as humans. Xuedong Huang, the company's chief speech scientist, reports that in a recent benchmark evaluation against the industry standard Switchboard speech recognition task, Microsoft researchers achieved a word error rate (WER) of 6.3 percent, the lowest in the industry. In a research paper published Tuesday, the scientists said: "Our best single system achieves an error rate of 6.9% on the NIST 2000 Switchboard set. We believe this is the best performance reported to date for a recognition system not based on system combination. This past weekend, at Interspeech, an international conference on speech communication and technology held in San Francisco, IBM said it has achieved a WER of 6.6 percent. Twenty years ago, the error rate of the best published research system had a WER of greater than 43 percent. "This new milestone benefited from a wide range of new technologies ...