Artificial intelligence (AI) is no longer the stuff of science fiction. The technology is already disrupting multiple industries, many of which impact you on a daily basis. Own an iPhone X? Its facial recognition system is powered by AI. Ever been redirected by Google Maps because of an accident or construction ahead? And those are just a couple of small examples.

A recent line of work focused on making adversarial training computationally efficient for deep learning models. In particular, Wong et al. (2020) showed that $\ell_\infty$-adversarial training with fast gradient sign method (FGSM) can fail due to a phenomenon called "catastrophic overfitting", when the model quickly loses its robustness over a single epoch of training. We show that adding a random step to FGSM, as proposed in Wong et al. (2020), does not prevent catastrophic overfitting, and that randomness is not important per se -- its main role being simply to reduce the magnitude of the perturbation. Moreover, we show that catastrophic overfitting is not inherent to deep and overparametrized networks, but can occur in a single-layer convolutional network with a few filters. In an extreme case, even a single filter can make the network highly non-linear locally, which is the main reason why FGSM training fails. Based on this observation, we propose a new regularization method, GradAlign, that prevents catastrophic overfitting by explicitly maximizing the gradient alignment inside the perturbation set and improves the quality of the FGSM solution. As a result, GradAlign allows to successfully apply FGSM training also for larger $\ell_\infty$-perturbations and reduce the gap to multi-step adversarial training. The code of our experiments is available at https://github.com/tml-epfl/understanding-fast-adv-training.

There is an old saying, "If rubber is the material that opened the way to the ground, aluminum is the one that opened the way to the sky." New materials were always discovered at each turning point that changed human history. Materials used in memory devices are also drastically evolving with the emergence of new materials such as doped silicon materials, resistance changing materials, and materials that spontaneously magnetize and polarize. How are these new materials made? A research team from POSTECH has revealed the mechanism behind making materials used in new memory devices by using artificial intelligence.

We design and train a neural network (NN) model to efficiently predict the infrared spectra of interstellar polycyclic aromatic hydrocarbons (PAHs) with a computational cost many orders of magnitude lower than what a first-principles calculation would demand. The input to the NN is based on the Morgan fingerprints extracted from the skeletal formulas of the molecules and does not require precise geometrical information such as interatomic distances. The model shows excellent predictive skill for out-of-sample inputs, making it suitable for improving the mixture models currently used for understanding the chemical composition and evolution of the interstellar medium. We also identify the constraints to its applicability caused by the limited diversity of the training data and estimate the prediction errors using a ensemble of NNs trained on subsets of the data. The power of these topological descriptors is demonstrated by the limited effect of including detailed geometrical information in the form of Coulomb matrix eigenvalues.

Mining frequent episodes aims at recovering sequential patterns from temporal data sequences, which can then be used to predict the occurrence of related events in advance. On the other hand, gradual patterns that capture co-variation of complex attributes in the form of " when X increases/decreases, Y increases/decreases" play an important role in many real world applications where huge volumes of complex numerical data must be handled. Recently, these patterns have received attention from the data mining community exploring temporal data who proposed methods to automatically extract gradual patterns from temporal data. However, to the best of our knowledge, no method has been proposed to extract gradual patterns that regularly appear at identical time intervals in many sequences of temporal data, despite the fact that such patterns may add knowledge to certain applications, such as e-commerce. In this paper, we propose to extract co-variations of periodically repeating attributes from the sequences of temporal data that we call seasonal gradual patterns. For this purpose, we formulate the task of mining seasonal gradual patterns as the problem of mining periodic patterns in multiple sequences and then we exploit periodic pattern mining algorithms to extract seasonal gradual patterns. We discuss specific features of these patterns and propose an approach for their extraction based on mining periodic frequent patterns common to multiple sequences. We also propose a new anti-monotonous support definition associated to these seasonal gradual patterns. The illustrative results obtained from some real world data sets show that the proposed approach is efficient and that it can extract small sets of patterns by filtering numerous nonseasonal patterns to identify the seasonal ones.

The chemical industry is -without question- one of the most important industries in the world. Not only do 90% of our everyday products contain chemicals, but the industry also employs approximately 10 million people. Naturally, they were one of the first to embrace digital technologies such as process control systems or sensors which have a long tradition in production. According to Frithjof Netzer, Senior Vice-President and Project Lead 4.0 of BASF: "A lot of energy and momentum in the field of digital can be observed, Chemicals are catching up. It is not the question if, but rather what and how it will be done." A continuous digital transformation plays a crucial role in several key aspects of the industry.

Learning to optimize (L2O) has gained increasing attention since classical optimizers require laborious problem-specific design and hyperparameter tuning. However, there is a gap between the practical demand and the achievable performance of existing L2O models. Specifically, those learned optimizers are applicable to only a limited class of problems, and often exhibit instability. With many efforts devoted to designing more sophisticated L2O models, we argue for another orthogonal, under-explored theme: the training techniques for those L2O models. We show that even the simplest L2O model could have been trained much better. We first present a progressive training scheme to gradually increase the optimizer unroll length, to mitigate a well-known L2O dilemma of truncation bias (shorter unrolling) versus gradient explosion (longer unrolling). We further leverage off-policy imitation learning to guide the L2O learning, by taking reference to the behavior of analytical optimizers. Our improved training techniques are plugged into a variety of state-of-the-art L2O models, and immediately boost their performance, without making any change to their model structures. Especially, by our proposed techniques, an earliest and simplest L2O model can be trained to outperform the latest complicated L2O models on a number of tasks. Our results demonstrate a greater potential of L2O yet to be unleashed, and urge to rethink the recent progress. Our codes are publicly available at: https://github.com/VITA-Group/L2O-Training-Techniques.

The fundamental goal of generative drug design is to propose optimized molecules that meet predefined activity, selectivity, and pharmacokinetic criteria. Despite recent progress, we argue that existing generative methods are limited in their ability to favourably shift the distributions of molecular properties during optimization. We instead propose a novel Reinforcement Learning framework for molecular design in which an agent learns to directly optimize through a space of synthetically-accessible drug-like molecules. This becomes possible by defining transitions in our Markov Decision Process as chemical reactions, and allows us to leverage synthetic routes as an inductive bias. We validate our method by demonstrating that it outperforms existing state-of the art approaches in the optimization of pharmacologically-relevant objectives, while results on multi-objective optimization tasks suggest increased scalability to realistic pharmaceutical design problems.

Aspen Technology, Inc. recently announced the general availability of the aspenONE V12 software release, which embeds artificial intelligence (AI) across the portfolio, and uses the cloud for delivery of enterprise-wide analytics and insights for increased safety, sustainability, and improved margins. AspenTech's Industrial AI solutions democratize the application of AI where it can deliver most value and is a vital step towards the Self-Optimizing Plant. "Aspen Hybrid Models are a major advance in the field of chemical engineering. Hybrid models are a major step forward in bringing together AspenTech's process models and machine learning and are a game changer in process engineering and plant improvement," said Dr. Karuna Potdar, Vice President, Technology Centre of Excellence, Reliance Industries Limited. Aspen Hybrid Models capture data from assets across the enterprise, and then apply AI, engineering first principles and AspenTech's domain expertise to deliver comprehensive, more accurate models at enterprise speed and scale.

However, it has even more dire consequences for those suffering from malnutrition and hunger.