White matter hyperintensity (WMH) is commonly found in elder individuals and appears to be associated with brain diseases. U-net is a convolutional network that has been widely used for biomedical image segmentation. Recently, U-net has been successfully applied to WMH segmentation. Random initialization is usally used to initialize the model weights in the U-net. However, the model may coverage to different local optima with different randomly initialized weights. We find a combination of thresholding and averaging the outputs of U-nets with different random initializations can largely improve the WMH segmentation accuracy. Based on this observation, we propose a post-processing technique concerning the way how averaging and thresholding are conducted. Specifically, we first transfer the score maps from three U-nets to binary masks via thresholding and then average those binary masks to obtain the final WMH segmentation. Both quantitative analysis (via the Dice similarity coefficient) and qualitative analysis (via visual examinations) reveal the superior performance of the proposed method. This post-processing technique is independent of the model used. As such, it can also be applied to situations where other deep learning models are employed, especially when random initialization is adopted and pre-training is unavailable.

This paper examines to what degree current deep learning architectures for image caption generation capture spatial language. On the basis of the evaluation of examples of generated captions from the literature we argue that systems capture what objects are in the image data but not where these objects are located: the captions generated by these systems are the output of a language model conditioned on the output of an object detector that cannot capture fine-grained location information. Although language models provide useful knowledge for image captions, we argue that deep learning image captioning architectures should also model geometric relations between objects.

Natural language processing (NLP) can be done using either top-down (theory driven) and bottom-up (data driven) approaches, which we call mechanistic and phenomenological respectively. The approaches are frequently considered to stand in opposition to each other. Examining some recent approaches in deep learning we argue that deep neural networks incorporate both perspectives and, furthermore, that leveraging this aspect of deep learning may help in solving complex problems within language technology, such as modelling language and perception in the domain of spatial cognition.

Deep reinforcement learning, and especially the Asynchronous Advantage Actor-Critic algorithm, has been successfully used to achieve super-human performance in a variety of video games. Starcraft II is a new challenge for the reinforcement learning community with the release of pysc2 learning environment proposed by Google Deepmind and Blizzard Entertainment. Despite being a target for several AI developers, few have achieved human level performance. In this project we explain the complexities of this environment and discuss the results from our experiments on the environment. We have compared various architectures and have proved that transfer learning can be an effective paradigm in reinforcement learning research for complex scenarios requiring skill transfer.

Deep Learning is one of the newest trends in Machine Learning and Artificial Intelligence research. It is also one of the most popular scientific research trends now-a-days. Deep learning methods have brought revolutionary advances in computer vision and machine learning. Every now and then, new and new deep learning techniques are being born, outperforming state-of-the-art machine learning and even existing deep learning techniques. In recent years, the world has seen many major breakthroughs in this field. Since deep learning is evolving at a huge speed, its kind of hard to keep track of the regular advances especially for new researchers. In this paper, we are going to briefly discuss about recent advances in Deep Learning for past few years.

Taxi destination prediction is a very important task for optimizing the efficiency of electronic dispatching systems, thus allowing relevant advantages for both taxi companies and customers. In fact, during periods of high demand, there should be a taxi whose current ride will end near a requested pick up location from a new customer. If an electronic dispatcher is able to know in advance where all taxi drivers will end their current ride, it will also be able to better allocate its resources, identifying which taxi to assign to each call. Moreover, automatic systems for the taxi mobility monitoring collect data that, integrated with other information sources, can help in understanding daytime human mobility routines. In this paper, we introduce a novel approach for addressing the taxi destination prediction problem, based on Recurrent Neural Networks (RNNs) applied to a regression setting. RNNs are trained based on the individual drivers' history and on geographical information (i.e., points of interest), using only the starting point of each ride (with no knowledge about the whole trajectory). The proposed approach was tested on the dataset of the ECML/PKDD Discovery Challenge 2015 - based on the city of Porto - obtaining better results with respect to the competition winner, whilst using less information, and on Manhattan and San Francisco datasets.

Each iteration of the learning was captured then later complied into a video, meaning we can actually see the learning process on the neural net frame by frame.

One week into my Research Science role at Lyft, I merged my first pull request into the Fraud team's code repository and deployed our fraud decision service. No, it wasn't to launch a groundbreaking user behavior activity-based convolutional recurrent neural network trained in a semi-supervised, adversarial fashion that challenges a user to prove her identity -- it would be a couple of years before that. Embarrassingly, it was to remove a duplicate line of feature coefficients in a hand-coded logistic regression model rolled out a little less than a year before. This small bug exposed a number of limitations of a system built primarily for a different type of usage -- that of business rules that encapsulate simple, human-readable handcrafted logic. In our old worldview, models were simply extensions of business rules.

Over the past few years, adversarial examples have received a significant amount of attention in the deep learning community. In this blog post, we want to share our high-level perspective on this phenomenon and how it fits into a larger question of robustness in machine learning. In subsequent posts, we plan to delve deeper into the topics that we will only briefly touch on today. To set the stage for our discussion, let us briefly introduce adversarial examples. On the left, we have an image of a pig that is correctly classified as such by a state-of-the-art convolutional neural network. After perturbing the image slightly (every pixel is in the range [0, 1] and changed by at most 0.005), the network now returns class "airliner" with high confidence.

It's infinite potential waiting to be tapped. With cognitive capabilities--like artificial intelligence, image recognition, or deep learning--data can help you see in all directions and get to smarter, surer insights swiftly.