Archive

Modern web-based social networks have become platforms where individuals can express personal views and discuss relevant issues in real-time. The possibility of analysing this massive aggregation of thoughts and opinions has applications in several domains, ranging from finance and marketing to the social sciences. However, the development of social media analysis tools is still slow and expensive, as most of the current approaches depend on...

In some imaging modalities based on coherent radiation, the noise contaminating an image may contain useful information, thereby necessitating the separation of the noise field rather than just denoising. When the algebraic operation that relates the image and noise is known, the noise component can be estimated in a straightforward manner after denoising. For truly multiplicative noise, such as the Rayleigh, Gamma, and other noises,...

In this talk, we address the problem of placing sensor probes in the brain such that the system dynamics’ are generically observable. The system dynamics whose states can encode for instance the fire-rating of the neurons or their ensemble following a neural-topological (structural) approach, and the sensors are assumed to be dedicated, i.e., can only measure a state at each time. Even though the mathematical...

We present a neural network model that computes embeddings of words using recurrent network based on long short-term memories to read in characters. As an alternative to word lookup tables that require a set of parameters for every word type in the vocabulary, our models only require a look up table for characters and a fixed number of parameters for the compositional model, independent of...

This talk addresses the problem of parameter identification for switched ARX system. The identification of such systems typically results in non-convex optimization problems, where finding the globally optimal solution exhibits exponential computational complexity in the size of the input. The exponential complexity may however not be tractable even for middle size problems. Another approach involves heuristics in order to reduce the computational complexity, with the...

In the last few years, language modeling techniques based on exponential models have consistently outperformed traditional n-gram models. Such techniques include L1-Regularized Maximum Entropy (L1-MaxEnt), and both Feedforward and Recurrent Neural Network Language Models (RNNLM). While more accurate, these models are also much more expensive to train and use. This presents a problem for low latency applications where it is desirable to find n-gram approximations...

This talk considers the recently introduced ordered weighted L1 (OWL) regularizer for sparse estimation problems with correlated variables. We begin by reviewing several convex analysis results concerning the OWL regularizer, namely: that it is indeed a norm, its dual norm, efficient methods to compute the corresponding proximity operator and the Euclidean projection on an OWL ball. We will also show how the OWL norm can...

Jet engines are complex machines, subject to demanding operating conditions, whose proper maintenance is critical to ensure high safety, maximum availability and minimum lifecycle costs. Therefore, dozens of sensors measure in real-time hundreds of engine parameters (temperature, pressure, vibration, shaft rotation speeds…) to monitor and predict the deterioration of jet engines. We present some of the results obtained during a PhD thesis held at the Instituto...

Unbabel is a startup whose mission is to enable seamless, trustworthy written communication in different languages. The Unbabel platform combines a novel approach to machine translation with a community of bilinguals and freelance translators which results in human quality translations, at a fraction of the cost, an order of magnitude faster. This talk will focus on the Unbabel translation pipeline and on the different challenges in tries...

Hyperspectral cameras acquire electromagnetic energy scattered in their instantaneous field view in hundreds or thousands of spectral channels with higher spectral resolution than multispectral cameras. This enhanced spectral resolution enables material identification via spectroscopic analysis, which facilitates countless applications that require identifying materials in scenarios unsuitable for classical spectroscopic analysis. However, due to low spatial resolution of hyperspectral cameras, microscopic material mixing, and multiple scattering,...