Autoregressive language models pretrained on large corpora have been successful at solving downstream tasks, even with zero-shot usage. However, there is little theoretical justification for their success. This paper considers the following questions: (1) Why should learning the distribution of natural language help with downstream classification tasks? (2) Why do features learned using language modeling help solve downstream tasks with linear classifiers? For (1), we hypothesize, and verify empirically, that classification tasks of interest can be reformulated as next word prediction tasks, thus making language modeling a meaningful pretraining task. For (2), we analyze properties of the cross-entropy objective to show that $\epsilon$-optimal language models in cross-entropy (log-perplexity) learn features that are $\mathcal{O}(\sqrt{\epsilon})$-good on natural linear classification tasks, thus demonstrating mathematically that doing well on language modeling can be beneficial for downstream tasks. We perform experiments to verify assumptions and validate theoretical results. Our theoretical insights motivate a simple alternative to the cross-entropy objective that performs well on some linear classification tasks.

Deep neural networks can achieve great successes when presented with large data sets and sufficient computational resources. However, their ability to learn new concepts quickly is quite limited. Meta-learning is one approach to address this issue, by enabling the network to learn how to learn. The exciting field of Deep Meta-Learning advances at great speed, but lacks a unified, insightful overview of current techniques. This work presents just that. After providing the reader with a theoretical foundation, we investigate and summarize key methods, which are categorized into i) metric-, ii) model-, and iii) optimization-based techniques. In addition, we identify the main open challenges, such as performance evaluations on heterogeneous benchmarks, and reduction of the computational costs of meta-learning.

For many reasoning-heavy tasks, it is challenging to find an appropriate end-to-end differentiable approximation to domain-specific inference mechanisms. Neural-Symbolic (NeSy) AI divides the end-to-end pipeline into neural perception and symbolic reasoning, which can directly exploit general domain knowledge such as algorithms and logic rules. However, it suffers from the exponential computational complexity caused by the interface between the two components, where the neural model lacks direct supervision, and the symbolic model lacks accurate input facts. As a result, they usually focus on learning the neural model with a sound and complete symbolic knowledge base while avoiding a crucial problem: where does the knowledge come from? In this paper, we present Abductive Meta-Interpretive Learning ($Meta_{Abd}$), which unites abduction and induction to learn perceptual neural network and first-order logic theories simultaneously from raw data. Given the same amount of domain knowledge, we demonstrate that $Meta_{Abd}$ not only outperforms the compared end-to-end models in predictive accuracy and data efficiency but also induces logic programs that can be re-used as background knowledge in subsequent learning tasks. To the best of our knowledge, $Meta_{Abd}$ is the first system that can jointly learn neural networks and recursive first-order logic theories with predicate invention.

We present a method for learning representations of entities, that uses a Transformer-based architecture as an entity encoder, and link prediction training on a knowledge graph with textual entity descriptions. We demonstrate that our approach can be applied effectively for link prediction in different inductive settings involving entities not seen during training, outperforming related state-of-the-art methods (22% MRR improvement on average). We provide evidence that the learned representations transfer to other tasks that do not require fine-tuning the entity encoder. In an entity classification task we obtain an average improvement of 16% accuracy compared with baselines that also employ pre-trained models. For an information retrieval task, significant improvements of up to 8.8% in NDCG@10 were obtained for natural language queries.

In the field of machine learning there is a growing interest towards more robust and generalizable algorithms. This is for example important to bridge the gap between the environment in which the training data was collected and the environment where the algorithm is deployed. Machine learning algorithms have increasingly been shown to excel in finding patterns and correlations from data. Determining the consistency of these patterns and for example the distinction between causal correlations and nonsensical spurious relations has proven to be much more difficult. In this paper a regularization scheme is introduced that prefers universal causal correlations. This approach is based on 1) the robustness of causal correlations and 2) the data not being independently and identically distribute (i.i.d.). The scheme is demonstrated with a classification task by clustering the (non-i.i.d.) training set in subpopulations. A non-i.i.d. regularization term is then introduced that penalizes weights that are not invariant over these clusters. The resulting algorithm favours correlations that are universal over the subpopulations and indeed a better performance is obtained on an out-of-distribution test set with respect to a more conventional l_2-regularization.

Autonomous agents optimize the reward function we give them. What they don't know is how hard it is for us to design a reward function that actually captures what we want. When designing the reward, we might think of some specific training scenarios, and make sure that the reward will lead to the right behavior in those scenarios. Inevitably, agents encounter new scenarios (e.g., new types of terrain) where optimizing that same reward may lead to undesired behavior. Our insight is that reward functions are merely observations about what the designer actually wants, and that they should be interpreted in the context in which they were designed. We introduce inverse reward design (IRD) as the problem of inferring the true objective based on the designed reward and the training MDP. We introduce approximate methods for solving IRD problems, and use their solution to plan risk-averse behavior in test MDPs. Empirical results suggest that this approach can help alleviate negative side effects of misspecified reward functions and mitigate reward hacking.

Regression is a modeling task that involves predicting a numerical value given an input. Algorithms used for regression tasks are also referred to as "regression" algorithms, with the most widely known and perhaps most successful being linear regression. Linear regression fits a line or hyperplane that best describes the linear relationship between inputs and the target numeric value. If the data contains outlier values, the line can become biased, resulting in worse predictive performance. Robust regression refers to a suite of algorithms that are robust in the presence of outliers in training data.

Kernel methods, a new generation of learning algorithms, utilize techniques from optimization, statistics, and functional analysis to achieve maximal generality, flexibility, and performance. These algorithms are different from earlier techniques used in machine learning in many respects: For example, they are explicitly based on a theoretical model of learning rather than on loose analogies with natural learning systems or other heuristics. They come with theoretical guarantees about their performance and have a modular design that makes it possible to separately implement and analyze their components. They are not affected by the problem of local minima because their training amounts to convex optimization. In the last decade, a sizable community of theoreticians and practitioners has formed around these methods, and a number of practical applications have been realized.

Diabetes management is a difficult task for patients, who must monitor and control their blood glucose levels in order to avoid serious diabetic complications. It is a difficult task for physicians, who must manually interpret large volumes of blood glucose data to tailor therapy to the needs of each patient. This paper describes three emerging applications that employ AI to ease this task: (1) case-based decision support for diabetes management; (2) machine learning classification of blood glucose plots; and (3) support vector regression for blood glucose prediction. The first application provides decision support by detecting blood glucose control problems and recommending therapeutic adjustments to correct them. The second provides an automated screen for excessive glycemic variability.

Support Vector Machine (SVM) is a widely-used supervised machine learning algorithm. It is mostly used in classification tasks but suitable for regression tasks as well. In this post, we dive deep into two important hyperparameters of SVMs, C and gamma, and explain their effects with visualizations. So I will assume you have a basic understanding of the algorithm and focus on these hyperparameters. SVM separates data points that belong to different classes with a decision boundary.