Sensory learning in the mammalian cortex has long been hypothesized to involve the objective of variational inference (VI). Likely the most well-known algorithm for cortical VI is the Wake-Sleep algorithm (Hinton et al. 1995). However Wake-Sleep problematically assumes that neural activities are independent given lower-layers during generation. Here, we construct a VI system that is both compatible with neurobiology and avoids this assumption. The core of the system is a wake-sleep discriminator that classifies network states as inferred or self-generated. Inference connections learn by opposing this discriminator. This adversarial dynamic solves a core problem within VI, which is to match the distribution of stimulus-evoked (inference) activity to that of self-generated activity. Meanwhile, generative connections learn to predict lower-level activity as in standard VI. We implement this algorithm and show that it can successfully train the approximate inference network for generative models. Our proposed algorithm makes several biological predictions that can be tested. Most importantly, it predicts a teaching signal that is remarkably similar to known properties of the cholinergic system.

Normalizing flows have shown great promise for modelling flexible probability distributions in a computationally tractable way. However, whilst data is often naturally described on Riemannian manifolds such as spheres, torii, and hyperbolic spaces, most normalizing flows implicitly assume a flat geometry, making them either misspecified or ill-suited in these situations. To overcome this problem, we introduce Riemannian continuous normalizing flows, a model which admits the parametrization of flexible probability measures on smooth manifolds by defining flows as the solution to ordinary differential equations. We show that this approach can lead to substantial improvements on both synthetic and real-world data when compared to standard flows or previously introduced projected flows.

In recent years, surrogate models have been successfully used in likelihood-free inference to decrease the number of simulator evaluations. The current state-of-the-art performance for this task has been achieved by Bayesian Optimization with Gaussian Processes (GPs). While this combination works well for unimodal target distributions, it is restricting the flexibility and applicability of Bayesian Optimization for accelerating likelihood-free inference more generally. We address this problem by proposing a Deep Gaussian Process (DGP) surrogate model that can handle more irregularly behaved target distributions. Our experiments show how DGPs can outperform GPs on objective functions with multimodal distributions and maintain a comparable performance in unimodal cases. This confirms that DGPs as surrogate models can extend the applicability of Bayesian Optimization for likelihood-free inference (BOLFI), while adding computational overhead that remains negligible for computationally intensive simulators.

Learning the distribution of observed data is a basic task in unsupervised learning which has been studied for decades. The recently-introduced concept of Generative Adversarial Networks (GANs) [1] has demonstrated great success in various distribution learning tasks. Unlike the traditional maximum-likelihood-based approaches, GANs learn the distribution of observed data through a zero-sum game between two machine players, a generator G mimicking the true distribution of data and a discriminator D distinguishing the generator's produced samples from real data points. This zero-sum game is typically formulated through a minimax optimization problem where G and D optimize a minimax objective quantifying how dissimilar G's generated samples and real training samples are. In GAN minimax optimization problems, the generator and discriminator functions are commonly chosen as two deep neural networks (DNNs). Leveraging the expressive power of DNNs, GANs have achieved state-of-the-art performance in learning complex distributions of image data [2, 3, 4]. This success, however, is achieved at the cost of their notoriously difficult training procedure which has introduced several challenges to the machine learning community. Addressing these challenges requires a deeper theoretical understanding of GANs, including their approximation, generalization, and optimization properties. Specifically, GANs have been frequently observed to fail in learning multi-modal distributions [5].

In most of the transfer learning approaches to reinforcement learning (RL) the distribution over the tasks is assumed to be stationary. Therefore, the target and source tasks are i.i.d. samples of the same distribution. In the context of this work, we consider the problem of transferring value functions through a variational method when the distribution that generates the tasks is time-variant, proposing a solution that leverages this temporal structure inherent in the task generating process. Furthermore, by means of a finite-sample analysis, the previously mentioned solution is theoretically compared to its time-invariant version. Finally, we will provide an experimental evaluation of the proposed technique with three distinct temporal dynamics in three different RL environments.

Ising machines (IMs) are a potential breakthrough in the NP-hard problem of score-based Bayesian network (BN) learning. To utilize the power of IMs, encoding of BN learning into quadratic unconstrained binary optimization (QUBO) has been proposed using up to $\mathcal{O}(N^2)$ bits, for $N$ variables in BN and $M = 2$ parents each. However, this approach is usually infeasible owing to the upper bound of IM bits when $M \geq 3$. In this paper, we propose an efficient conversion method for BN learning into QUBO with a maximum of $\sum_n (\Lambda_n - 1) + \binom N2$ bits, for $\Lambda_n$ parent set candidates each. The advance selection of parent set candidates plays an essential role in reducing the number of required bits. We also develop a pre-processing algorithm based on the capabilities of a classification and regression tree (CART), which allows us to search for parent set candidates consistent with score minimization in a realistic timeframe.Our conversion method enables us to more significantly reduce the upper bound of the required bits in comparison to an existing method, and is therefore expected to make a significant contribution to the advancement of scalable score-based BN learning.

Computers have been used to analyze and create music since they were first introduced in the 1950s and 1960s. Beginning in the late 1990s, the rise of the Internet and large scale platforms for music recommendation and retrieval have made music an increasingly prevalent domain of machine learning and artificial intelligence research. While still nascent, several different approaches have been employed to tackle what may broadly be referred to as "musical intelligence." This article provides a definition of musical intelligence, introduces a taxonomy of its constituent components, and surveys the wide range of AI methods that can be, and have been, brought to bear in its pursuit, with a particular emphasis on machine learning methods.

The unification of logic and probability is a long-standing concern in AI, and more generally, in the philosophy of science. In essence, logic provides an easy way to specify properties that must hold in every possible world, and probability allows us to further quantify the weight and ratio of the worlds that must satisfy a property. To that end, numerous developments have been undertaken, culminating in proposals such as probabilistic relational models. While this progress has been notable, a general-purpose first-order knowledge representation language to reason about probabilities and dynamics, including in continuous settings, is still to emerge. In this paper, we survey recent results pertaining to the integration of logic, probability and actions in the situation calculus, which is arguably one of the oldest and most well-known formalisms. We then explore reduction theorems and programming interfaces for the language. These results are motivated in the context of cognitive robotics (as envisioned by Reiter and his colleagues) for the sake of concreteness. Overall, the advantage of proving results for such a general language is that it becomes possible to adapt them to any special-purpose fragment, including but not limited to popular probabilistic relational models.

The goal of item response theoretic (IRT) models is to provide estimates of latent traits from binary observed indicators and at the same time to learn the item response functions (IRFs) that map from latent trait to observed response. However, in many cases observed behavior can deviate significantly from the parametric assumptions of traditional IRT models. Nonparametric IRT models overcome these challenges by relaxing assumptions about the form of the IRFs, but standard tools are unable to simultaneously estimate flexible IRFs and recover ability estimates for respondents. We propose a Bayesian nonparametric model that solves this problem by placing Gaussian process priors on the latent functions defining the IRFs. This allows us to simultaneously relax assumptions about the shape of the IRFs while preserving the ability to estimate latent traits. This in turn allows us to easily extend the model to further tasks such as active learning. GPIRT therefore provides a simple and intuitive solution to several longstanding problems in the IRT literature.

Classifying sequential data as early as and as accurately as possible is a challenging yet critical problem, especially when a sampling cost is high. One algorithm that achieves this goal is the sequential probability ratio test (SPRT), which is known as Bayes-optimal: it can keep the expected number of data samples as small as possible, given the desired error upper-bound. The SPRT has recently been found to be the best model that explains the activities of the neurons in the primate parietal cortex that are thought to mediate our complex decision-making processes. However, the original SPRT makes two critical assumptions that limit its application in real-world scenarios: (i) samples are independently and identically distributed, and (ii) the likelihood of the data being derived from each class can be calculated precisely. Here, we propose the SPRT-TANDEM, a deep neural network-based SPRT algorithm that overcomes the above two obstacles. The SPRT-TANDEM estimates the log-likelihood ratio of two alternative hypotheses by leveraging a novel Loss function for Log-Likelihood Ratio estimation (LLLR), while allowing for correlations up to $N (\in \mathbb{N})$ preceding samples. In tests on one original and two public video databases, Nosaic MNIST, UCF101, and SiW, the SPRT-TANDEM achieves statistically significantly better classification accuracy than other baseline classifiers, with a smaller number of data samples. The code and Nosaic MNIST are publicly available at https://github.com/TaikiMiyagawa/SPRT-TANDEM.