The reparameterization trick has become one of the most useful tools in the field of variational inference. However, the reparameterization trick is based on the standardization transformation which restricts the scope of application of this method to distributions that have tractable inverse cumulative distribution functions or are expressible as deterministic transformations of such distributions. In this paper, we generalized the reparameterization trick by allowing a general transformation. We discover that the proposed model is a special case of control variate indicating that the proposed model can combine the advantages of CV and generalized reparameterization. Based on the proposed gradient model, we propose a new polynomial-based gradient estimator which has better theoretical performance than the reparameterization trick under certain condition and can be applied to a larger class of variational distributions. In studies of synthetic and real data, we show that our proposed gradient estimator has a significantly lower gradient variance than other state-of-the-art methods thus enabling a faster inference procedure.

We propose an algorithm for inexpensive gradient-based hyperparameter optimization that combines the implicit function theorem (IFT) with efficient inverse Hessian approximations. We present results about the relationship between the IFT and differentiating through optimization, motivating our algorithm. We use the proposed approach to train modern network architectures with millions of weights and millions of hyper-parameters. For example, we learn a data-augmentation network - where every weight is a hyperparameter tuned for validation performance - outputting augmented training examples. Jointly tuning weights and hyperparameters with our approach is only a few times more costly in memory and compute than standard training.

Deep neural networks frequently contain far more weights, represented at a higher precision, than are required for the specific task which they are trained to perform. Consequently, they can often be compressed using techniques such as weight pruning and quantization that reduce both model size and inference time without appreciable loss in accuracy. Compressing models before they are deployed can therefore result in significantly more efficient systems. However, while the results are desirable, finding the best compression strategy for a given neural network, target platform, and optimization objective often requires extensive experimentation. Moreover, finding optimal hyperparameters for a given compression strategy typically results in even more expensive, frequently manual, trial-and-error exploration. In this paper, we introduce a programmable system for model compression called Condensa. Users programmatically compose simple operators, in Python, to build complex compression strategies. Given a strategy and a user-provided objective, such as minimization of running time, Condensa uses a novel sample-efficient constrained Bayesian optimization algorithm to automatically infer desirable sparsity ratios. Our experiments on three real-world image classification and language modeling tasks demonstrate memory footprint reductions of up to 65x and runtime throughput improvements of up to 2.22x using at most 10 samples per search. We have released a reference implementation of Condensa at https://github.com/NVlabs/condensa.

In this paper, the problem of energy efficient transmission and computation resource allocation for federated learning (FL) over wireless communication networks is investigated. In the considered model, each user exploits limited local computational resources to train a local FL model with its collected data and, then, sends the trained FL model parameters to a base station (BS) which aggregates the local FL model and broadcasts it back to all of the users. Since FL involves an exchange of a learning model between users and the BS, both computation and communication latencies are determined by the learning accuracy level. Meanwhile, due to the limited energy budget of the wireless users, both local computation energy and transmission energy must be considered during the FL process. This joint learning and communication problem is formulated as an optimization problem whose goal is to minimize a weighted sum of the completion time of FL, local computation energy, and transmission energy of all users, that captures the tradeoff of latency and energy consumption for FL. To solve this problem, an iterative algorithm is proposed where, at every step, closed-form solutions for time allocation, bandwidth allocation, power control, computation frequency, and learning accuracy are derived. For the special case that only minimizes the completion time, a bisection-based algorithm is proposed to obtain the optimal solution. Numerical results show that the proposed algorithms can reduce up to 25.6% delay and 37.6% energy consumption compared to conventional FL methods.

Outlier detection is a core task in data mining with a plethora of algorithms that have enjoyed wide scale usage. Existing algorithms are primarily focused on detection, that is the identification of outliers in a given dataset. In this paper we explore the relatively under-studied problem of the outlier explanation problem. Our goal is, given a dataset that is already divided into outliers and normal instances, explain what characterizes the outliers. We explore the novel direction of a semantic explanation that a domain expert or policy maker is able to understand. We formulate this as an optimization problem to find explanations that are both interpretable and pure. Through experiments on real-world data sets, we quantitatively show that our method can efficiently generate better explanations compared with rule-based learners.

In this paper we propose the first correct poly-time algorithm for exact partitioning of high-order models (a worst case NP-hard problem). We define a general class of $m$-degree Homogeneous Polynomial Models, which subsumes several examples motivated from prior literature. Exact partitioning can be formulated as a tensor optimization problem. We relax this NP-hard problem to a convex conic form problem (poly-time solvable by interior point methods). To this end, we carefully define the positive semidefinite tensor cone, and show its convexity, and the convexity of its dual cone. This allows us to construct a primal-dual certificate to show that the solution of the convex relaxation is correct (equal to the unobserved true group assignment) under some sample complexity conditions.

Optimization is becoming increasingly common in scientific and engineering domains. Oftentimes, these problems involve various levels of stochasticity or uncertainty in generating proposed solutions. Therefore, optimization in these scenarios must consider this stochasticity to properly guide the design of future experiments. Here, we adapt Bayesian optimization to handle uncertain outcomes, proposing a new framework called stochastic sampling Bayesian optimization (SSBO). We show that the bounds on expected regret for an upper confidence bound search in SSBO resemble those of earlier Bayesian optimization approaches, with added penalties due to the stochastic generation of inputs. Additionally, we adapt existing batch optimization techniques to properly limit the myopic decision making that can arise when selecting multiple instances before feedback. Finally, we show that SSBO techniques properly optimize a set of standard optimization problems as well as an applied problem inspired by bioengineering.

Stochastic gradient descent (SGD) is the workhorse of modern machine learning. Sometimes, there are many different potential gradient estimators that can be used. When so, choosing the one with the best tradeoff between cost and variance is important. This paper analyzes the convergence rates of SGD as a function of time, rather than iterations. This results in a simple rule to select the estimator that leads to the best optimization convergence guarantee. This choice is the same for different variants of SGD, and with different assumptions about the objective (e.g. convexity or smoothness). Inspired by this principle, we propose a technique to automatically select an estimator when a finite pool of estimators is given. Then, we extend to infinite pools of estimators, where each one is indexed by control variate weights. This is enabled by a reduction to a mixed-integer quadratic program. Empirically, automatically choosing an estimator performs comparably to the best estimator chosen with hindsight.

In variational inference, the complicated target is approximated by a closest member relative to the Kullback-Leibler (KL) divergence in a pre-specified family of tractable densities. In many large-scale machine learning applications including clustering problems [11, 32], image classification [25, 27] and topic models [21, 7], variational inference can be orders of magnitude faster than the traditional sampling based approaches such as Markov Chain Monte Carlo (MCMC). In particular, by turning the integration, or sampling, problem into an optimization problem, variational inference can take advantage of modern optimization tools such as stochastic optimization techniques [20, 17] and distributed optimization architecture [1, 8] for further improving its efficiency. Among various approximating schemes, mean-field approximation is the most common type of variational inference that is conceptually simple, implementation-wise easy and particularly suitable for problems involving large numbers of latent variables. The word "mean-field" is originated from the mean-field theory in physics where despite complex interactions among many particles in a many (infinite) body system, all interactions to any one particle can be approximated by a single averaged effect from a "mean-field". In variational inference, by restricting the approximating family of the mean-field to be all density functions that are fully factorized over (blocks of) unknown variables, the associated optimization problem of finding a closest weih2@illinois.edu

Unsupervised text embedding has shown great power in a wide range of NLP tasks. While text embeddings are typically learned in the Euclidean space, directional similarity is often more effective in tasks such as word similarity and document clustering, which creates a gap between the training stage and usage stage of text embedding. To close this gap, we propose a spherical generative model based on which unsupervised word and paragraph embeddings are jointly learned. To learn text embeddings in the spherical space, we develop an efficient optimization algorithm with convergence guarantee based on Riemannian optimization. Our model enjoys high efficiency and achieves state-of-the-art performances on various text embedding tasks including word similarity and document clustering.