Current deep neural networks are highly overparameterized (up to billions of connection weights) and nonlinear. Yet they can fit data almost perfectly through variants of gradient descent algorithms and achieve unexpected levels of prediction accuracy without overfitting. These are formidable results that escape the bias-variance predictions of statistical learning and pose conceptual challenges for non-convex optimization. In this paper, we use methods from statistical physics of disordered systems to analytically study the computational fallout of overparameterization in nonconvex neural network models. As the number of connection weights increases, we follow the changes of the geometrical structure of different minima of the error loss function and relate them to learning and generalisation performance. We find that there exist a gap between the SAT/UNSAT interpolation transition where solutions begin to exist and the point where algorithms start to find solutions, i.e. where accessible solutions appear. This second phase transition coincides with the discontinuous appearance of atypical solutions that are locally extremely entropic, i.e., flat regions of the weight space that are particularly solution-dense and have good generalization properties. Although exponentially rare compared to typical solutions (which are narrower and extremely difficult to sample), entropic solutions are accessible to the algorithms used in learning. We can characterize the generalization error of different solutions and optimize the Bayesian prediction, for data generated from a structurally different network. Numerical tests on observables suggested by the theory confirm that the scenario extends to realistic deep networks.

We say an algorithm is batch size-invariant if changes to the batch size can largely be compensated for by changes to other hyperparameters. Stochastic gradient descent is well-known to have this property at small batch sizes, via the learning rate. However, some policy optimization algorithms (such as PPO) do not have this property, because of how they control the size of policy updates. In this work we show how to make these algorithms batch size-invariant. Our key insight is to decouple the proximal policy (used for controlling policy updates) from the behavior policy (used for off-policy corrections). Our experiments help explain why these algorithms work, and additionally show how they can make more efficient use of stale data.

Two-level stochastic optimization formulations have become instrumental in a number of machine learning contexts such as neural architecture search, continual learning, adversarial learning, and hyperparameter tuning. Practical stochastic bilevel optimization problems become challenging in optimization or learning scenarios where the number of variables is high or there are constraints. The goal of this paper is twofold. First, we aim at promoting the use of bilevel optimization in large-scale learning and we introduce a practical bilevel stochastic gradient method (BSG-1) that requires neither lower level second-order derivatives nor system solves (and dismisses any matrix-vector products). Our BSG-1 method is close to first-order principles, which allows it to achieve a performance better than those that are not, such as DARTS. Second, we develop bilevel stochastic gradient descent for bilevel problems with lower level constraints, and we introduce a convergence theory that covers the unconstrained and constrained cases and abstracts as much as possible from the specifics of the bilevel gradient calculation.

Welcome to the ninth update from the Gradient! If you were referred by a friend, subscribe and follow us on Twitter! This news edition's story is Sharing learnings from the first algorithmic bias bounty challenge. Summary Twitter's algorithmic bias bounty challenge, the first of its kind, recently concluded. While users had previously found the algorithm had a racial bias, the bounty uncovered a number of other biases and potential harms.

Optimization of discrete structures aims at generating a new structure with the better property given an existing one, which is a fundamental problem in machine learning. Different from the continuous optimization, the realistic applications of discrete optimization (e.g., text generation) are very challenging due to the complex and long-range constraints, including both syntax and semantics, in discrete structures. In this work, we present SAGS, a novel Simulated Annealing framework for Graph and Sequence optimization. The key idea is to integrate powerful neural networks into metaheuristics (e.g., simulated annealing, SA) to restrict the search space in discrete optimization. We start by defining a sophisticated objective function, involving the property of interest and pre-defined constraints (e.g., grammar validity). SAGS searches from the discrete space towards this objective by performing a sequence of local edits, where deep generative neural networks propose the editing content and thus can control the quality of editing. We evaluate SAGS on paraphrase generation and molecule generation for sequence optimization and graph optimization, respectively. Extensive results show that our approach achieves state-of-the-art performance compared with existing paraphrase generation methods in terms of both automatic and human evaluations. Further, SAGS also significantly outperforms all the previous methods in molecule generation.

Rule-based models, e.g., decision trees, are widely used in scenarios demanding high model interpretability for their transparent inner structures and good model expressivity. However, rule-based models are hard to optimize, especially on large data sets, due to their discrete parameters and structures. Ensemble methods and fuzzy/soft rules are commonly used to improve performance, but they sacrifice the model interpretability. To obtain both good scalability and interpretability, we propose a new classifier, named Rule-based Representation Learner (RRL), that automatically learns interpretable non-fuzzy rules for data representation and classification. To train the non-differentiable RRL effectively, we project it to a continuous space and propose a novel training method, called Gradient Grafting, that can directly optimize the discrete model using gradient descent. An improved design of logical activation functions is also devised to increase the scalability of RRL and enable it to discretize the continuous features end-to-end. Exhaustive experiments on nine small and four large data sets show that RRL outperforms the competitive interpretable approaches and can be easily adjusted to obtain a trade-off between classification accuracy and model complexity for different scenarios. Our code is available at: https://github.com/12wang3/rrl.

I had my first encounter with the Gradient Descent algorithm when I was learning about Linear Regression for the very first time. I devoured information about Gradient Descent as much as I could. Scouring through the internet looking for an explanation that would satisfy me. The most common explanation I found was analogous to the "going downhill on a cliff" experience. While this was really intuitive and easily comprehensible.

Several methods have been proposed to solve such derivative-free stochastic optimization problems, and we refer the reader to [3, 38] for surveys of these methods. A popular class of these methods estimate the gradients using function values and employ standard gradient-based optimization methods using these estimators. Quasi-Newton methods are recognized as one of the most powerful methods for solving deterministic optimization problems. These methods build quadratic models of the objective information using only gradient information. Recently, researchers have been adapting these methods for stochastic settings when the gradient information is available. The empirical results in [15] indicate that a careful implementation of these methods can be efficient compared with the popular stochastic gradient methods. We adapt these methods to make them suitable for situations where the gradients are estimated using function values. We propose finite-difference derivative-free stochastic quasi-Newton methods for solving (1) by exploiting common random number (CRN) evaluations of f.

Deception is a crucial tool in the cyberdefence repertoire, enabling defenders to leverage their informational advantage to reduce the likelihood of successful attacks. One way deception can be employed is through obscuring, or masking, some of the information about how systems are configured, increasing attacker's uncertainty about their targets. We present a novel game-theoretic model of the resulting defender-attacker interaction, where the defender chooses a subset of attributes to mask, while the attacker responds by choosing an exploit to execute. The strategies of both players have combinatorial structure with complex informational dependencies, and therefore even representing these strategies is not trivial. First, we show that the problem of computing an equilibrium of the resulting zero-sum defender-attacker game can be represented as a linear program with a combinatorial number of system configuration variables and constraints, and develop a constraint generation approach for solving this problem. Next, we present a novel highly scalable approach for approximately solving such games by representing the strategies of both players as neural networks. The key idea is to represent the defender's mixed strategy using a deep neural network generator, and then using alternating gradient-descent-ascent algorithm, analogous to the training of Generative Adversarial Networks. Our experiments, as well as a case study, demonstrate the efficacy of the proposed approach.

This blog introduces different types of multi classification systems. Multiclass classifiers can distinguish between more than two classes other than binary classifiers. Stochastic gradient descent (SGD) classifiers, Random Forest classifiers, and naive Bayes classifiers etc. are capable of handling multiple classes natively. On the other hand, Logistic Regression or Support Vector Machine classifiers are strictly binary classifiers. There are various strategies that you can use to perform multiclass classification with multiple binary classifiers.