Su, Jiahao, Byeon, Wonmin, Huang, Furong, Kautz, Jan, Anandkumar, Animashree

Higher-order Recurrent Neural Networks (RNNs) are effective for long-term forecasting since such architectures can model higher-order correlations and long-term dynamics more effectively. However, higher-order models are expensive and require exponentially more parameters and operations compared with their first-order counterparts. This problem is particularly pronounced in multidimensional data such as videos. To address this issue, we propose Convolutional Tensor-Train Decomposition (CTTD), a novel tensor decomposition with convolutional operations. With CTTD, we construct Convolutional Tensor-Train LSTM (Conv-TT-LSTM) to capture higher-order space-time correlations in videos. We demonstrate that the proposed model outperforms the conventional (first-order) Convolutional LSTM (ConvLSTM) as well as the state-of-the-art ConvLSTM-based approaches in pixel-level video prediction tasks on Moving-MNIST and KTH action datasets, but with much fewer parameters.

Atz, Kenneth, Grisoni, Francesca, Schneider, Gisbert

Geometric deep learning (GDL), which is based on neural network architectures that incorporate and process symmetry information, has emerged as a recent paradigm in artificial intelligence. GDL bears particular promise in molecular modeling applications, in which various molecular representations with different symmetry properties and levels of abstraction exist. This review provides a structured and harmonized overview of molecular GDL, highlighting its applications in drug discovery, chemical synthesis prediction, and quantum chemistry. Emphasis is placed on the relevance of the learned molecular features and their complementarity to well-established molecular descriptors. This review provides an overview of current challenges and opportunities, and presents a forecast of the future of GDL for molecular sciences.

Udrescu, Silviu-Marian, Tegmark, Max

We present a method for unsupervised learning of equations of motion for objects in raw and optionally distorted unlabeled video. We first train an autoencoder that maps each video frame into a low-dimensional latent space where the laws of motion are as simple as possible, by minimizing a combination of non-linearity, acceleration and prediction error. Differential equations describing the motion are then discovered using Pareto-optimal symbolic regression. We find that our pre-regression ("pregression") step is able to rediscover Cartesian coordinates of unlabeled moving objects even when the video is distorted by a generalized lens. Using intuition from multidimensional knot-theory, we find that the pregression step is facilitated by first adding extra latent space dimensions to avoid topological problems during training and then removing these extra dimensions via principal component analysis.

Decision-making under uncertainty is hugely important for any decisions sensitive to perturbations in observed data. One method of incorporating uncertainty into making optimal decisions is through robust optimization, which minimizes the worst-case scenario over some uncertainty set. We explore Mahalanobis distance as a novel function for multi-target regression and the construction of joint prediction regions. We also connect conformal prediction regions to robust optimization, providing finite sample valid and conservative uncertainty sets, aptly named conformal uncertainty sets. We compare the coverage and efficiency of the conformal prediction regions generated with Mahalanobis distance to other conformal prediction regions. We also construct a small robust optimization example to compare conformal uncertainty sets to those constructed under the assumption of normality.

Kuchibhotla, Arun K., Ramdas, Aaditya K.

We provide an alternate unified framework for conformal prediction, which is a framework to provide assumption-free prediction intervals. Instead of beginning by choosing a conformity score, our framework starts with a sequence of nested sets $\{\mathcal{F}_t(x)\}_{t\in\mathcal{T}}$ for some ordered set $\mathcal{T}$ that specifies all potential prediction sets. We show that most proposed conformity scores in the literature, including several based on quantiles, straightforwardly result in nested families. Then, we argue that what conformal prediction does is find a mapping $\alpha \mapsto t(\alpha)$, meaning that it calibrates or rescales $\mathcal{T}$ to $[0,1]$. Nestedness is a natural and intuitive requirement because the optimal prediction sets (eg: level sets of conditional densities) are also nested, but we also formally prove that nested sets are universal, meaning that any conformal prediction method can be represented in our framework. Finally, to demonstrate its utility, we show how to develop the full conformal, split conformal, cross-conformal and the recent jackknife+ methods within our nested framework, thus immediately generalizing the latter two classes of methods to new settings. Specifically, we prove the validity of the leave-one-out, $K$-fold, subsampling and bootstrap variants of the latter two methods for any nested family.