Dimitar Filev on bringing cutting-edge computational intelligence to cars and the factories that build them. Driverless cars aren't the only application for deep learning on the road: neural networks have begun to make their way into every corner of the automotive industry, from supply-chain management to engine controllers. In this installment of our ongoing series on artificial intelligence (AI) and machine learning (ML) in the enterprise, we speak with Dimitar Filev, executive technical leader at Ford Research & Advanced Engineering, who leads the team focused on control methods and computational intelligence. Ford research lab has been conducting systematic research on computational intelligence--one of the branches of AI--for more than 20 years. About 15 years ago, Ford Motor Company introduced one of the first large-scale industrial applications of neural networks.

The Internet of things (IoT) is the inter-networking of physical devices (also termed as connected devices or smart devices), vehicles, buildings and other objects (which could be smart wearable, diagnostic device, kitchen appliances etc.) embedded with electronics, software, sensors, actuators, and network connectivity that enables these "smart objects" to collect and exchange data. In other words, Internet of things is a global infrastructure for the information society. IoT allows advanced services by interconnecting (physical and virtual) things based on existing and evolving interoperable information and communication technologies. For example, the smart refrigerator in your kitchen (at home) can send you an alert (or notification) on your smartphone (while you are leaving office) when you're out of milk or gas. Your wearable or smart watch can warn you if there is something wrong with your pulse or heart-rate. Additionally, all these information gets recorded.

When artificial intelligence (AI) is discussed today, most people are referring to machine learning algorithms or deep learning systems. While AI has advanced significantly over the years, the principle behind these technologies remains the same. Someone trains a system to receive certain data and asks it to produce a specified outcome -- it's up to the machine to develop its own algorithm to reach this outcome. Alas, while we've been able to create some very smart systems, they are not foolproof. Data science competition platform Kaggle wants to prepare AI systems for super-smart cyberattacks, and they're doing so by pitting AI against AI in a contest dubbed the Competition on Adversarial Attacks and Defenses.

Counting objects in digital images is a process that should be replaced by machines. This tedious task is time consuming and prone to errors due to fatigue of human annotators. The goal is to have a system that takes as input an image and returns a count of the objects inside and justification for the prediction in the form of object localization. We repose a problem, originally posed by Lempitsky and Zisserman, to instead predict a count map which contains redundant counts based on the receptive field of a smaller regression network. The regression network predicts a count of the objects that exist inside this frame. By processing the image in a fully convolutional way each pixel is going to be accounted for some number of times, the number of windows which include it, which is the size of each window, (i.e., 32x32 = 1024). To recover the true count we take the average over the redundant predictions. Our contribution is redundant counting instead of predicting a density map in order to average over errors. We also propose a novel deep neural network architecture adapted from the Inception family of networks called the Count-ception network. Together our approach results in a 20% relative improvement (2.9 to 2.3 MAE) over the state of the art method by Xie, Noble, and Zisserman in 2016.

We propose an approach to build a neural machine translation system with no supervised resources (i.e., no parallel corpora) using multimodal embedded representation over texts and images. Based on the assumption that text documents are often likely to be described with other multimedia information (e.g., images) somewhat related to the content, we try to indirectly estimate the relevance between two languages. Using multimedia as the "pivot", we project all modalities into one common hidden space where samples belonging to similar semantic concepts should come close to each other, whatever the observed space of each sample is. This modality-agnostic representation is the key to bridging the gap between different modalities. Putting a decoder on top of it, our network can flexibly draw the outputs from any input modality. Notably, in the testing phase, we need only source language texts as the input for translation. In experiments, we tested our method on two benchmarks to show that it can achieve reasonable translation performance. We compared and investigated several possible implementations and found that an end-to-end model that simultaneously optimized both rank loss in multimodal encoders and cross-entropy loss in decoders performed the best.

Given that in practice training data is scarce for all but a small set of problems, a core question is how to incorporate prior knowledge into a model. In this paper, we consider the case of prior procedural knowledge for neural networks, such as knowing how a program should traverse a sequence, but not what local actions should be performed at each step. To this end, we present an end-to-end differentiable interpreter for the programming language Forth which enables programmers to write program sketches with slots that can be filled with behaviour trained from program input-output data. We can optimise this behaviour directly through gradient descent techniques on user-specified objectives, and also integrate the program into any larger neural computation graph. We show empirically that our interpreter is able to effectively leverage different levels of prior program structure and learn complex behaviours such as sequence sorting and addition. When connected to outputs of an LSTM and trained jointly, our interpreter achieves state-of-the-art accuracy for end-to-end reasoning about quantities expressed in natural language stories.

Continuing the tour of older papers that started with our ResNet blog post, we now take on Skip-Thought Vectors by Kiros et al. Their goal was to come up with a useful embedding for sentences that was not tuned for a single task and did not require labeled data to train. They took inspiration from Word2Vec skip-gram (you can find my explanation of that algorithm here) and attempt to extend it to sentences. Skip-thought vectors are created using an encoder-decoder model. The encoder takes in the training sentence and outputs a vector.

Google's AI boutique, DeepMind, known for dispelling human delusions of intellectual superiority by soundly beating the world's top Go players with computer code, has found that instilling its software agents with something like imagination helps them learn better. In two papers published this week – "Imagination-Augmented Agents for Deep Reinforcement Learning" and "Learning model-based planning from scratch" – the AI biz's brain boffins, based in Britain, describe novel techniques for improving deep reinforcement learning through what can generously be described as imaginative planning. Reinforcement learning is a form of machine learning. It involves a software agent that learns by interacting with a specific environment, usually through trial and error. Deep learning is a form of machine that involves algorithms inspired by the human brain, called neural networks.

I recently stumbled across an old Data Science Stack Exchange answer of mine on the topic of the "Best Python library for neural networks", and it struck me how much the Python deep learning ecosystem has evolved over the course of the past 2.5 years. The library I recommended in July 2014,, is no longer actively developed or maintained, but a whole host of deep learning libraries have sprung up to take its place. Each has its own strengths and weaknesses. We've used most of the technologies on this list in production or development at indico, but for the few that we haven't, I'll pull from the experiences of others to help give a clear, comprehensive picture of the Python deep learning ecosystem of 2017. In particular, we'll be looking at: Theano is a Python library that allows you to define, optimize, and evaluate mathematical expressions involving multi-dimensional arrays efficiently.

System makers Fujitsu and Huawei Technologies reportedly are both planning to develop processors optimized for artificial intelligence workloads, moves that will put them into competition with the likes of Intel, Google, Nvidia and Advanced Micro Devices. Tech vendors are pushing hard to bring artificial intelligence (AI) and deep learning capabilities into their portfolios to meet the growing demand generated by a broad range of workloads, from data analytics to self-driving vehicles. Microsoft, Google, IBM and others are creating AI business units and building out products and services that can leverage the technologies. Chip makers also are making the move. Intel last week unveiled its latest generation Xeon server chips that, among other improvements, deliver 2.2 times the performance for deep learning training and inference tasks than their predecessors.