Transfer learning aims at improving the performance of target learners on target domains by transferring the knowledge contained in different but related source domains. In this way, the dependence on a large number of target domain data can be reduced for constructing target learners. Due to the wide application prospects, transfer learning has become a popular and promising area in machine learning. Although there are already some valuable and impressive surveys on transfer learning, these surveys introduce approaches in a relatively isolated way and lack the recent advances in transfer learning. As the rapid expansion of the transfer learning area, it is both necessary and challenging to comprehensively review the relevant studies. This survey attempts to connect and systematize the existing transfer learning researches, as well as to summarize and interpret the mechanisms and the strategies in a comprehensive way, which may help readers have a better understanding of the current research status and ideas. Different from previous surveys, this survey paper reviews over forty representative transfer learning approaches from the perspectives of data and model. The applications of transfer learning are also briefly introduced. In order to show the performance of different transfer learning models, twenty representative transfer learning models are used for experiments. The models are performed on three different datasets, i.e., Amazon Reviews, Reuters-21578, and Office-31. And the experimental results demonstrate the importance of selecting appropriate transfer learning models for different applications in practice.

Water temperature is known to be principal driver of the growth, survival, and reproduction of economically viable fish [21, 30] (see Appendix for more details). Increases in water temperature are also linked to the occurrence of aquatic invasive species [28, 29], which may displace fish and native aquatic organisms, and further result in harmful algal blooms [9, 26]. Hence, accurate and timely information about water temperature is necessary to monitor the ecological health of lakes and forecast future populations of fish and other aquatic taxa. Since observations of water temperatures are incomplete at broad spatial scales (or nonexistent for most lakes), physics-based models of lake temperature, e.g., the General Lake Model (GLM) [10], are commonly used for studying lake processes. A standard formulation in these models is to assume horizontal heterogeneity is limited and that the most relevant dynamics are captured in the vertical dimension of the lake, thereby modeling the lake as a series of vertical layers. These modeling studies often use temperature of water at the centre of a lake at varying depth values 1 and time points for model validation. We adopt the same formulation to model the temperature of water in a lake, Y d,tat depth d and time t . In particular, we leverage two key physical principles of our problem to guide neural network approaches, briefly described in the following.

Retinal vessel segmentation is an essential step for fundus image analysis. With the recent advances of deep learning technologies, many convolutional neural networks have been applied in this field, including the successful U-Net. In this work, we firstly modify the U-Net with functional blocks aiming to pursue higher performance. The absence of the expected performance boost then lead us to dig into the opposite direction of shrinking the U-Net and exploring the extreme conditions such that its segmentation performance is maintained. Experiment series to simplify the network structure, reduce the network size and restrict the training conditions are designed. Results show that for retinal vessel segmentation on DRIVE database, U-Net does not degenerate until surprisingly acute conditions: one level, one filter in convolutional layers, and one training sample. This experimental discovery is both counter-intuitive and worthwhile. Not only are the extremes of the U-Net explored on a well-studied application, but also one intriguing warning is raised for the research methodology which seeks for marginal performance enhancement regardless of the resource cost.

Programming by example is the problem of synthesizing a program from a small set of input / output pairs. Recent works applying machine learning methods to this task show promise, but are typically reliant on generating synthetic examples for training. A particular challenge lies in generating meaningful sets of inputs and outputs, which well-characterize a given program and accurately demonstrate its behavior. Where examples used for testing are generated by the same method as training data then the performance of a model may be partly reliant on this similarity. In this paper we introduce a novel approach using an SMT solver to synthesize inputs which cover a diverse set of behaviors for a given program. We carry out a case study comparing this method to existing synthetic data generation procedures in the literature, and find that data generated using our approach improves both the discriminatory power of example sets and the ability of trained machine learning models to generalize to unfamiliar data.

Travel time estimation is a fundamental problem in transportation science with extensive literature. The study of these techniques has intensified due to availability of many publicly available large trip datasets. Recently developed deep learning based models have improved the generality and performance and have focused on estimating times for individual sub-trajectories and aggregating them to predict the travel time of the entire trajectory. However, these techniques ignore the road network information. In this work, we propose and study techniques for incorporating road networks along with historical trips' data into travel time prediction. We incorporate both node embeddings as well as road distance into the existing model. Experiments on large real-world benchmark datasets suggest improved performance, especially when the train data is small. As expected, the proposed method performs better than the baseline when there is a larger difference between road distance and Vincenty distance between start and end points.

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.

Machine-learning (ML) hardware and software system demand is burgeoning. Driven by ML applications, the number of different ML inference systems has exploded. Over 100 organizations are building ML inference chips, and the systems that incorporate existing models span at least three orders of magnitude in power consumption and four orders of magnitude in performance; they range from embedded devices to data-center solutions. Fueling the hardware are a dozen or more software frameworks and libraries. The myriad combinations of ML hardware and ML software make assessing ML-system performance in an architecture-neutral, representative, and reproducible manner challenging. There is a clear need for industry-wide standard ML benchmarking and evaluation criteria. MLPerf Inference answers that call. Driven by more than 30 organizations as well as more than 200 ML engineers and practitioners, MLPerf implements a set of rules and practices to ensure comparability across systems with wildly differing architectures. In this paper, we present the method and design principles of the initial MLPerf Inference release. The first call for submissions garnered more than 600 inference-performance measurements from 14 organizations, representing over 30 systems that show a range of capabilities.

--T uning machine learning models at scale, especially finding the right hyperparameter values, can be difficult and time-consuming. In addition to the computational effort required, this process also requires some ancillary efforts including engineering tasks (e.g., job scheduling) as well as more mundane tasks (e.g., keeping track of the various parameters and associated results). We present Auptimizer, a general Hyperparameter Optimization (HPO) framework to help data scientists speed up model tuning and bookkeeping. With Auptimizer, users can use all available computing resources in distributed settings for model training. The design also allows researchers to integrate new HPO algorithms. T o demonstrate its flexibility, we show how Auptimizer integrates a few major HPO techniques (from random search to neural architecture search). Designing a Machine Learning (ML) framework for production faces challenges similar to those faced with Big Data. There is a large volume of models with a variety of configurations and training them efficiently at scale with reproducibility is critical to realizing their business value. In this paper, we address one design aspect of the ML framework, namely the HPO process, via a framework called Auptimizer. A. Hyperparameter Optimization ML models are typically sensitive to the values of hy-perparameters [31]. Different from model parameters, these hyperparameters are values that control the model configuration or the training setup and thus need to be set before training the model. Due to the lack of gradient information for these hyperparameters, tuning them is often treated as a black-box optimization [11]. As an alternative to manual selection (which is usually based on modeler's expertise), researchers have proposed different methods to accelerate the tuning process including Bayesian approaches [26], evolutionary algorithms [10], multi-armed bandits [8], and architecture search by learning [33]. Tuning hyperparameters is often time-consuming especially when model training is computationally intensive [1]. Therefore, in practice, an automated HPO solution is critically important for machine learning. Both open-source solutions and commercial offerings are available.

--Data parallelism has become the de facto standard for training Deep Neural Network on multiple processing units. In this work we propose DC-S3GD, a decentralized (without Parameter Server) stale-synchronous version of the Delay-Compensated Asynchronous Stochastic Gradient Descent (DC-ASGD) algorithm. In our approach, we allow for the overlap of computation and communication, and compensate the inherent error with a first-order correction of the gradients. We prove the effectiveness of our approach by training Convolutional Neural Network with large batches and achieving state-of- the-art results. I NTRODUCTION Training Deep Neural Networks (DNNs) is a time-and resource-consuming problem. For example, to train a DNN to state-of-the-art accuracy on a single processing unit, the total time needed is in the order of magnitude of days, or even weeks [16]. For this reason, in recent years, several algorithms have been developed to allow users to perform parallel or distributed training of DNNs [7].

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.