Many natural optimization problems are NP-hard, which implies that they are probably hard to solve exactly in the worst-case. However, in practice, it suffices to get reasonably good solutions for all (or even most) instances. This paper presents a new algorithm for computing approximate solution in ${\Theta(N})$ for the MAX-E-$3$-SAT problem by using deep learning methodology. This methodology allows us to create a learning algorithm able to fix Boolean variables by using local information obtained by the Survey Propagation algorithm. By performing an accurate analysis, on random CNF instances of the MAX-E-$3$-SAT with several Boolean variables, we show that this new algorithm, avoiding any decimation strategy, can build assignments better than a random one, even if the convergence of the messages is not found. Although this algorithm is not competitive with state-of-the-art MAX-SAT solvers, it can solve substantially larger and more difficult problems than it ever saw during training.

This paper presents c2g-HOF networks which learn to generate cost-to-go functions for manipulator motion planning. The c2g-HOF architecture consists of a cost-to-go function over the configuration space represented as a neural network (c2g-network) as well as a Higher Order Function (HOF) network which outputs the weights of the c2g-network for a given input workspace. Both networks are trained end-to-end in a supervised fashion using costs computed from traditional motion planners. Once trained, c2g-HOF can generate a smooth and continuous cost-to-go function directly from workspace sensor inputs (represented as a point cloud in 3D or an image in 2D). At inference time, the weights of the c2g-network are computed very efficiently and near-optimal trajectories are generated by simply following the gradient of the cost-to-go function. We compare c2g-HOF with traditional planning algorithms for various robots and planning scenarios. The experimental results indicate that planning with c2g-HOF is significantly faster than other motion planning algorithms, resulting in orders of magnitude improvement when including collision checking. Furthermore, despite being trained from sparsely sampled trajectories in configuration space, c2g-HOF generalizes to generate smoother, and often lower cost, trajectories. We demonstrate cost-to-go based planning on a 7 DoF manipulator arm where motion planning in a complex workspace requires only 0.13 seconds for the entire trajectory.

While cosmological dark matter-only simulations relying solely on gravitational effects are comparably fast to compute, baryonic properties in simulated galaxies require complex hydrodynamic simulations that are computationally costly to run. We explore the merging of an extended version of the equilibrium model, an analytic formalism describing the evolution of the stellar, gas, and metal content of galaxies, into a machine learning framework. In doing so, we are able to recover more properties than the analytic formalism alone can provide, creating a high-speed hydrodynamic simulation emulator that populates galactic dark matter haloes in N-body simulations with baryonic properties. While there exists a trade-off between the reached accuracy and the speed advantage this approach offers, our results outperform an approach using only machine learning for a subset of baryonic properties. We demonstrate that this novel hybrid system enables the fast completion of dark matter-only information by mimicking the properties of a full hydrodynamic suite to a reasonable degree, and discuss the advantages and disadvantages of hybrid versus machine learning-only frameworks. In doing so, we offer an acceleration of commonly deployed simulations in cosmology.

We've all heard of Kaggle, but that also means there's more competition -- recently, Kaggle reached 5 million users. Further, not all competitions are open to everyone in the world. "Members of the Kaggle community who are not United States Citizens or legal permanent residents at the time of entry are allowed to participate in the Competition but are not eligible to win prizes. If a team has one or more members who are not prize eligible, then the entire team is not prize eligible." By trying out other competition platforms, you can be a "big fish in a small pond," as there are a lot fewer competitors.

As healthcare professionals are facing massive pressure not only to ensure the quality of care, but also to come up with new solutions, cures and treatments, they are becoming increasingly dependent on advanced technologies like artificial intelligence (AI) and machine learning (ML). But it is hardly a smooth partnership. The issues of skills shortages at the entry-level and of "messy data" in leveraging patient records at the high end are merely book-ends for a range of challenges that span these fields. Last week's annual Amazon Web Services Re:Invent conference, one of the largest cloud-focused events in the world, saw the launch or demonstration of a range of new cloud-based tools that are ideal for health research and treatment. ML, defined as computer algorithms that improve automatically through experience, was at the heart of these.

It's been a busy few days at NeurIPS 2020 so far with all manner of talks, workshops, tutorials and socials on offer. This selection of tweets gives a flavour of the various events and discussions taking place. Go watch it right now, you won't regret it! Interesting talk by Chris Bishop at #NeurIPS2020 Basic or Applied research is not a 1D space. Next up at #NeurIPS2020: Shafi Goldwasser presenting on three works about privacy, verifiability, and robustness in machine learning.

Don't miss our final Global Cougs Speaker Series of the semester! Darlington Akogo, Founder of KaraAgro AI and minoHealth, will be speaking about the application of AI in African healthcare and agriculture. It will happen this Thursday, December 10th, at 10 a.m.

Just as the name implies, data science is a branch of science that applies the scientific method to data with the goal of studying the relationships between different features and drawing out meaningful conclusions based on these relationships. Data is, therefore, the key component in data science. A dataset is a particular instance of data that is used for analysis or model building at any given time. A dataset comes in different flavors such as numerical data, categorical data, text data, image data, voice data, and video data. A dataset could be static (not changing) or dynamic (changes with time, for example, stock prices).

Growing research demonstrates that synthetic failure modes imply poor generalization. We compare commonly used audio-to-audio losses on a synthetic benchmark, measuring the pitch distance between two stationary sinusoids. The results are surprising: many have poor sense of pitch direction. These shortcomings are exposed using simple rank assumptions. Our task is trivial for humans but difficult for these audio distances, suggesting significant progress can be made in self-supervised audio learning by improving current losses.

Gaussian Processes (GPs) are a class of kernel methods that have shown to be very useful in geoscience applications. They are widely used because they are simple, flexible and provide very accurate estimates for nonlinear problems, especially in parameter retrieval. An addition to a predictive mean function, GPs come equipped with a useful property: the predictive variance function which provides confidence intervals for the predictions. The GP formulation usually assumes that there is no input noise in the training and testing points, only in the observations. However, this is often not the case in Earth observation problems where an accurate assessment of the instrument error is usually available. In this paper, we showcase how the derivative of a GP model can be used to provide an analytical error propagation formulation and we analyze the predictive variance and the propagated error terms in a temperature prediction problem from infrared sounding data.