Are you a data analyst, data scientist, or a researcher looking for a guide that will help you increase the speed and efficiency of your machine learning activities? If yes, then this course is for you! Google's brainchild TensorFlow, in its first year, has more than 6000 open source repositories online. It has helped engineers, researchers, and many others make significant progress with everything from voice/sound recognition to language translation and face recognition. It has also proved to be useful in the early detection of skin cancer and preventing blindness in diabetics.

I saw that Fast.ai is shifting on PyTorch, I saw that PyTorch is utmost favourable for research prototyping. So, I decided to implement some research paper in PyTorch. I have already worked on C-DSSM model at Parallel Dots. But there my implementation was in Keras. I will emphasize on the hacker perspective, of porting the code from Keras to PyTorch, than the research perspective in the blog here.

In my previous blog post "How I stopped worrying and embraced docker microservices" I talked about why Microservices are the bees knees for scaling Machine Learning in production. A fair amount of time has passed (almost a year ago, whoa) and it proved that building Deep Learning pipelines in production is a more complex, multi-aspect problem. Yes, microservices are an amazing tool, both for software reuse, distributed systems design, quick failure and recovery, yada yada. But what seems very obvious now, is that Machine Learning services are very stateful, and statefulness is a problem for horizontal scaling. An easy way to deal with this issue is understand that ML models are large, and thus should not be context switched.

Essentially an architecture with a controller (FNN, RNN, LSTM) that has read/write access to a memory in order to learn a particular task. The controller gets an input, can read/write to memory block and then output a result. When reading and writing to the memory matrix, introspective attention is used. All of this is end-to-end differentiable, so theoretically we can learn very complex tasks.

In this paper, we focus on developing a novel mechanism to preserve differential privacy in deep neural networks, such that: (1) The privacy budget consumption is totally independent of the number of training steps; (2) It has the ability to adaptively inject noise into features based on the contribution of each to the output; and (3) It could be applied in a variety of different deep neural networks. To achieve this, we figure out a way to perturb affine transformations of neurons, and loss functions used in deep neural networks. In addition, our mechanism intentionally adds "more noise" into features which are "less relevant" to the model output, and vice-versa. Our theoretical analysis further derives the sensitivities and error bounds of our mechanism. Rigorous experiments conducted on MNIST and CIFAR-10 datasets show that our mechanism is highly effective and outperforms existing solutions.

Multi-Entity Dependence Learning (MEDL) explores conditional correlations among multiple entities. The availability of rich contextual information requires a nimble learning scheme that tightly integrates with deep neural networks and has the ability to capture correlation structures among exponentially many outcomes. We propose MEDL_CVAE, which encodes a conditional multivariate distribution as a generating process. As a result, the variational lower bound of the joint likelihood can be optimized via a conditional variational auto-encoder and trained end-to-end on GPUs. Our MEDL_CVAE was motivated by two real-world applications in computational sustainability: one studies the spatial correlation among multiple bird species using the eBird data and the other models multi-dimensional landscape composition and human footprint in the Amazon rainforest with satellite images. We show that MEDL_CVAE captures rich dependency structures, scales better than previous methods, and further improves on the joint likelihood taking advantage of very large datasets that are beyond the capacity of previous methods.

Abstract-- An accurate and reliable image based fruit detection system is critical for supporting higher level agriculture tasks such as yield mapping and robotic harvesting. This paper presents the use of a state-of-the-art object detection framework, Faster R-CNN, in the context of fruit detection in orchards, including mangoes, almonds and apples. Ablation studies are presented to better understand the practical deployment of the detection network, including how much training data is required to capture variability in the dataset. Data augmentation techniques are shown to yield significant performance gains, resulting in a greater than twofold reduction in the number of training images required. In contrast, transferring knowledge between orchards contributed to negligible performance gain over initialising the Deep Convolutional Neural Network directly from ImageNet features. Finally, to operate over orchard data containing between 100-1000 fruit per image, a tiling approach is introduced for the Faster R-CNN framework. The study has resulted in the best yet detection performance for these orchards relative to previous works, with an F1-score of 0.9 achieved for apples and mangoes. I. INTRODUCTION Vision based fruit detection is a critical component for infield automation in agriculture. With accurate knowledge of individual fruit locations in the field, it is possible to perform yield estimation and mapping, which is important for growers as it facilitates efficient utilisation of resources and improves returns per unit area and time. Precise localisation of the fruit is also a necessary component of an automated robotic harvesting system, which can help mitigate one of the most labour intensive tasks in an orchard [1].

We envision a world where devices, machines, automobiles, and things are much more intelligent, simplifying and enriching our daily lives. They will be able to perceive, reason, and take intuitive actions based on awareness of the situation, improving just about any experience and solving problems that to this point we've either left to the user, or to more conventional algorithms. Artificial intelligence (AI) is the technology driving this revolution. You may have heard this vision or may think that AI is really about big data and the cloud, and yet Qualcomm's solutions already have the power, thermal, and processing efficiency to run powerful AI algorithms on the actual device -- which brings several advantages. AI is a pervasive trend that is rapidly accelerating thanks to vast amounts of data and progress in both algorithms and the processing capacity of modern devices.

Neural network are powerful learning models especially deep learning networks on visual and speech recognition problems. This may result from their capacity of expressing arbitrary computation. However, it is still hard to fully understand their properties, thus, how they made the final decision after a sequence of decisions in a dynamic environment. Since we have many layers engaged in such complicated decision making process, we don't really understand how they think. In spite of having made a lot of efforts (e.g., a researcher created a popular toolkit called Deep Visualization Toolbox) to capture step by step how a neural network get trained, what we can see inside these layers is still very intricate.

However, for our current projects we need a compute server that we have exclusive access to." Access to a deep learning workstation will increase the speed of innovation and improve security." RESEARCHER "I felt I won the software stack lottery as NVIDIA- docker was already installed. I immediately pulled a container and started work on a CNTK NCCL project, the next day pulled another container to work on a TF biomedical project. I haven't looked back at how to reimage because felt too productive." It feels right for this work to allow fast iteration.