How can we enable machines to make sense of the world, and become better at learning? To approach this goal, I believe viewing intelligence in terms of many integral aspects, and also a universal two-term tradeoff between task performance and complexity, provides two feasible perspectives. In this thesis, I address several key questions in some aspects of intelligence, and study the phase transitions in the two-term tradeoff, using strategies and tools from physics and information. Firstly, how can we make the learning models more flexible and efficient, so that agents can learn quickly with fewer examples? Inspired by how physicists model the world, we introduce a paradigm and an AI Physicist agent for simultaneously learning many small specialized models (theories) and the domain they are accurate, which can then be simplified, unified and stored, facilitating few-shot learning in a continual way. Secondly, for representation learning, when can we learn a good representation, and how does learning depend on the structure of the dataset? We approach this question by studying phase transitions when tuning the tradeoff hyperparameter. In the information bottleneck, we theoretically show that these phase transitions are predictable and reveal structure in the relationships between the data, the model, the learned representation and the loss landscape. Thirdly, how can agents discover causality from observations? We address part of this question by introducing an algorithm that combines prediction and minimizing information from the input, for exploratory causal discovery from observational time series. Fourthly, to make models more robust to label noise, we introduce Rank Pruning, a robust algorithm for classification with noisy labels. I believe that building on the work of my thesis we will be one step closer to enable more intelligent machines that can make sense of the world.

Weakly supervised learning aims at coping with scarce labeled data. Previous weakly supervised studies typically assume that there is only one kind of weak supervision in data. In many applications, however, raw data usually contains more than one kind of weak supervision at the same time. For example, in user experience enhancement from Didi, one of the largest online ride-sharing platforms, the ride comment data contains severe label noise (due to the subjective factors of passengers) and severe label distribution bias (due to the sampling bias). We call such a problem as "compound weakly supervised learning". In this paper, we propose the CWSL method to address this problem based on Didi ride-sharing comment data. Specifically, an instance reweighting strategy is employed to cope with severe label noise in comment data, where the weights for harmful noisy instances are small. Robust criteria like AUC rather than accuracy and the validation performance are optimized for the correction of biased data label. Alternating optimization and stochastic gradient methods accelerate the optimization on large-scale data. Experiments on Didi ride-sharing comment data clearly validate the effectiveness. We hope this work may shed some light on applying weakly supervised learning to complex real situations.

This paper provides a systematic and comprehensive survey that reviews the latest research efforts focused on machine learning (ML) based performance improvement of wireless networks, while considering all layers of the protocol stack (PHY, MAC and network). First, the related work and paper contributions are discussed, followed by providing the necessary background on data-driven approaches and machine learning for non-machine learning experts to understand all discussed techniques. Then, a comprehensive review is presented on works employing ML-based approaches to optimize the wireless communication parameters settings to achieve improved network quality-of-service (QoS) and quality-of-experience (QoE). We first categorize these works into: radio analysis, MAC analysis and network prediction approaches, followed by subcategories within each. Finally, open challenges and broader perspectives are discussed.

An agreed upon definition of machine learning is, a computer program is said to have learned when it's performance measure P at task T improves with experience E. Under the definition of Supervised Learning, we get this diagram. Here the experience would be the training data required to improve the algorithm. In practice we put this data into the Design Matrix. Design Matrix [dəˈzīn ˈmātriks]: term -- if a single input can be represented as a vector, putting all of the training examples, i.e the vectors, into 1 matrix makes the entire input aspects of the training data. This is not all of the experience. We still need the labels, if the examples are the inputs.

Domain adaptation performance of a learning algorithm on a target domain is a function of its source domain error and a divergence measure between the data distribution of these two domains. We present a study of various distance-based measures in the context of NLP tasks, that characterize the dissimilarity between domains based on sample estimates. We first conduct analysis experiments to show which of these distance measures can best differentiate samples from same versus different domains, and are correlated with empirical results. Next, we develop a DistanceNet model which uses these distance measures, or a mixture of these distance measures, as an additional loss function to be minimized jointly with the task's loss function, so as to achieve better unsupervised domain adaptation. Finally, we extend this model to a novel DistanceNet-Bandit model, which employs a multi-armed bandit controller to dynamically switch between multiple source domains and allow the model to learn an optimal trajectory and mixture of domains for transfer to the low-resource target domain. We conduct experiments on popular sentiment analysis datasets with several diverse domains and show that our DistanceNet model, as well as its dynamic bandit variant, can outperform competitive baselines in the context of unsupervised domain adaptation. 1 Introduction In situations where large-scale annotated datasets are available, supervised learning algorithms have achieved remarkable progress in various NLP challenges (LeCun, Bengio, and Hinton 2015). Most supervised learning algorithms rely on the assumption that data distribution during training is the same as that during test. However, in many real-life scenarios, the data distribution of interest at test-time might be different from that during training. The process of collecting new datasets that reflect the new distribution is usually not scalable due to monetary as well as time constraints.

Ensemble learning is a very prevalent method employed in machine learning. The relative success of ensemble methods is attributed to its ability to tackle a wide range of instances and complex problems that require different low-level approaches. However, ensemble methods are relatively less popular in reinforcement learning owing to the high sample complexity and computational expense involved. We present a new training and evaluation framework for model-free algorithms that use ensembles of policies obtained from a single training instance. These policies are diverse in nature and are learned through directed perturbation of the model parameters at regular intervals.

Supervised learning solves modern analytics challenges and drives informed organizational decisions. Although the predictive power of machine learning models can be very impressive, there is no benefit unless they drive actions. Models must be deployed in an automated fashion to continually support decision making for measurable benefit. And while unsupervised methods open powerful analytic opportunities, they do not come with a clear path to deployment. This course will clarify when each approach best fits the business need and show you how to derive value from both approaches.

Semi-supervised learning aims to take advantage of a large amount of unlabeled data to improve the accuracy of a model that only has access to a small number of labeled examples. We propose curriculum labeling, an approach that exploits pseudo-labeling for propagating labels to unlabeled samples in an iterative and self-paced fashion. This approach is surprisingly simple and effective and surpasses or is comparable with the best methods proposed in the recent literature across all the standard benchmarks for image classification. Notably, we obtain 94.91% accuracy on CIFAR-10 using only 4,000 labeled samples, and 88.56% top-5 accuracy on Imagenet-ILSVRC using 128,000 labeled samples. In contrast to prior works, our approach shows improvements even in a more realistic scenario that leverages out-of-distribution unlabeled data samples.

Artificial intelligence (AI) has made astonishing progress in the last decade. AI can now drive cars, diagnose diseases from medical images, recommend movies, even whom you should date, make investment decisions, and create art that people have sold at auction. A lot of research today, however, focuses on teaching AI to do things the way we do them. For example, computer vision and natural language processing – two of the hottest research areas in the field – deal with building AI models that can see like humans and use language like humans. But instead of teaching computers to imitate human thought, the time has now come to let them evolve on their own, so instead of becoming like us, they have a chance to become better than us.

Structured prediction requires manipulating a large number of combinatorial structures, e.g., dependency trees or alignments, either as latent or output variables. Recently, the SparseMAP method has been proposed as a differentiable, sparse alternative to maximum a posteriori (MAP) and marginal inference. SparseMAP returns a combination of a small number of structures, a desirable property in some downstream applications. However, SparseMAP requires a tractable MAP inference oracle. This excludes, e.g., loopy graphical models or factor graphs with logic constraints, which generally require approximate inference. In this paper, we introduce LP-SparseMAP, an extension of SparseMAP that addresses this limitation via a local polytope relaxation. LP-SparseMAP uses the flexible and powerful domain specific language of factor graphs for defining and backpropagating through arbitrary hidden structure, supporting coarse decompositions, hard logic constraints, and higher-order correlations. We derive the forward and backward algorithms needed for using LP-SparseMAP as a hidden or output layer. Experiments in three structured prediction tasks show benefits compared to SparseMAP and Structured SVM.