We investigate the performance of machine learning algorithms trained exclusively with configurations obtained from importance sampling Monte Carlo simulations of the two-dimensional Ising model with conserved magnetization. For supervised machine learning, we use convolutional neural networks and find that the corresponding output not only allows to locate the phase transition point with high precision, it also displays a finite-size scaling characterized by an Ising critical exponent. For unsupervised learning, restricted Boltzmann machines (RBM) are trained to generate new configurations that are then used to compute various quantities. We find that RBM generates configurations with magnetizations and energies forbidden in the original physical system. The RBM generated configurations result in energy density probability distributions with incorrect weights as well as in wrong spatial correlations. We show that shortcomings are also encountered when training RBM with configurations obtained from the non-conserved Ising model.

We present extensive numerical simulations of a family of non-equilibrium Potts models with absorbing states that allows for a variety of scenarios, depending on the number of spin states and the range of the spin-spin interactions. These scenarios encompass a voter critical point, a discontinuous transition as well as the presence of both a symmetry-breaking phase transition and an absorbing phase transition. While we also investigate standard steady-state quantities, our emphasis is on time-dependent quantities that provide insights into the transient properties of the models.

We provide a construction for categorical representation learning and introduce the foundations of ‘categorifier’. The central theme in representation learning is the idea of everything to vector. Every object in a dataset S can be represented as a vector in R n by an encoding map E : O b j ( S ) → R n . More importantly, every morphism can be represented as a matrix E : H o m ( S ) → R n n . The encoding map E is generally modeled by a deep neural network. The goal of representation learning is to design appropriate tasks on the dataset to train the encoding map (assuming that an encoding is optimal if it universally optimizes the performance on various tasks). However, the latter is still a set-theoretic approach. The goal of the current article is to promote the representation learning to a new level via a category-theoretic approach. As a proof of concept, we provide an example of a text translator equipped with our technology, showing that our categorical learning model outperforms the current deep learning models by 17 times. The content of the current article is part of a US provisional patent application filed by QGNai, Inc.

Following the earlier formalism of the categorical representation learning, we discuss the construction of the ``RG-flow-based categorifier''. Borrowing ideas from the theory of renormalization group flows (RG) in quantum field theory, holographic duality, and hyperbolic geometry and combining them with neural ODE techniques, we construct a new algorithmic natural language processing (NLP) architecture, called the RG-flow categorifier or for short the RG categorifier, which is capable of data classification and generation in all layers. We apply our algorithmic platform to biomedical data sets and show its performance in the field of sequence-to-function mapping. In particular, we apply the RG categorifier to particular genomic sequences of flu viruses and show how our technology is capable of extracting the information from given genomic sequences, finding their hidden symmetries and dominant features, classifying them, and using the trained data to make a stochastic prediction of new plausible generated sequences associated with a new set of viruses which could avoid the human immune system.

Deep Neural Network (DNN) classifiers are known to be vulnerable to Trojan or backdoor attacks, where the classifier is manipulated such that it misclassifies any input containing an attacker-determined Trojan trigger. Backdoors compromise a model's integrity, thereby posing a severe threat to the landscape of DNN-based classification. While multiple defenses against such attacks exist for classifiers in the image domain, there have been limited efforts to protect classifiers in the text domain.We present Trojan-Miner (T-Miner) -a defense framework for Trojan attacks on DNN-based text classifiers. T-Miner employs a sequence-to-sequence (seq-2-seq) generative model that probes the suspicious classifier and learns to produce text sequences that are likely to contain the Trojan trigger. T-Miner then analyzes the text produced by the generative model to determine if they contain trigger phrases, and correspondingly, whether the tested classifier has a backdoor. T-Miner requires no access to the training dataset or clean inputs of the suspicious classifier, and instead uses synthetically crafted "nonsensical" text inputs to train the generative model. We extensively evaluate T-Miner on 1100 model instances spanning 3 ubiquitous DNN model architectures, 5 different classification tasks, and a variety of trigger phrases. We show that T-Miner detects Trojan and clean models with a 98.75% overall accuracy, while achieving low false positives on clean models. We also show that T-Miner is robust against a variety of targeted, advanced attacks from an adaptive attacker.