Imitation from Observation (IfO) prompts the robot to imitate tasks from unlabeled videos via reinforcement learning (RL). The performance of the IfO algorithm depends on its ability to extract task-relevant representations since images are informative. Existing IfO algorithms extract image representations by using a simple encoding network or pre-trained network. Due to the lack of action labels, it is challenging to design a supervised task-relevant proxy task to train the simple encoding network. Representations extracted by a pre-trained network such as Resnet are often task-irrelevant. In this article, we propose a new approach for robot IfO via multimodal observations. Different modalities describe the same information from different sides, which can be used to design an unsupervised proxy task. Our approach contains two modules: the unsupervised cross-modal representation (UCMR) module and a self-behavioral cloning (self-BC)-based RL module. The UCMR module learns to extract task-relevant representations via a multimodal unsupervised proxy task. The Self-BC for further offline policy optimization collects successful experiences during the RL training. We evaluate our approach on the real robot pouring water task, quantitative pouring task, and pouring sand task. The robot achieves state-of-the-art performance.

Unstructured data are a promising new source of information that insurance companies may use to understand their risk portfolio better and improve the customer experience. However, these novel data sources are difficult to incorporate into existing ratemaking frameworks due to the size and format of the unstructured data. This paper proposes a framework to use street view imagery within a generalized linear model. To do so, we use representation learning to extract an embedding vector containing useful information from the image. This embedding is dense and low dimensional, making it appropriate to use within existing ratemaking models. We find that there is useful information included in street view imagery to predict the frequency of claims for certain types of perils. This model can be used as in a ratemaking framework but also opens the door to future empirical research on attempting to extract which characteristics within the image leads to increased or decreased predicted claim frequencies. Throughout, we discuss the practical difficulties (technical and social) of using this type of data for insurance pricing.

We focus on modelling categorical features and improving predictive power of neural networks with mixed categorical and numerical features in supervised learning tasks. The goal of this paper is to challenge the current dominant approach in actuarial data science with a new architecture of a neural network and a new training algorithm. The key proposal is to use a joint embedding for all categorical features, instead of separate entity embeddings, to determine the numerical representation of the categorical features which is fed, together with all other numerical features, into hidden layers of a neural network with a target response. In addition, we postulate that we should initialize the numerical representation of the categorical features and other parameters of the hidden layers of the neural network with parameters trained with (denoising) autoencoders in unsupervised learning tasks, instead of using random initialization of parameters. Since autoencoders for categorical data play an important role in this research, they are investigated in more depth in the paper. We illustrate our ideas with experiments on a real data set with claim numbers, and we demonstrate that we can achieve a higher predictive power of the network.

Representation learning has proven to be a powerful methodology in a wide variety of machine-learning applications. For atmospheric dynamics, however, it has so far not been considered, arguably due to the lack of large-scale, labeled datasets that could be used for training. In this work, we show how to sidestep the difficulty and introduce a self-supervised learning task that is applicable to a wide variety of unlabeled atmospheric datasets. Specifically, we train a neural network on the simple yet intricate task of predicting the temporal distance between atmospheric fields from distinct but nearby times. We demonstrate that training with this task on the ERA5 reanalysis dataset leads to internal representations that capture intrinsic aspects of atmospheric dynamics. For example, when employed as a loss function in other machine-learning applications, the derived AtmoDist distance leads to improved results compared to the $ {\mathrm{\ell}}_2 $-loss. For downscaling one obtains higher resolution fields that match the true statistics more closely than previous approaches and for the interpolation of missing or occluded data the AtmoDist distance leads to results that contain more realistic fine-scale features. Since it is obtained from observational data, AtmoDist also provides a novel perspective on atmospheric predictability.

Spatial data are a rich source of information for actuarial applications: knowledge of a risk’s location could improve an insurance company’s ratemaking, reserving or risk management processes. Relying on historical geolocated loss data is problematic for areas where it is limited or unavailable. In this paper, we construct spatial embeddings within a complex convolutional neural network representation model using external census data and use them as inputs to a simple predictive model. Compared to spatial interpolation models, our approach leads to smaller predictive bias and reduced variance in most situations. This method also enables us to generate rates in territories with no historical experience.

Research on graph representation learning has received great attention in recent years since most data in real-world applications come in the form of graphs. High-dimensional graph data are often in irregular forms. They are more difficult to analyze than image/video/audio data defined on regular lattices. Various graph embedding techniques have been developed to convert the raw graph data into a low-dimensional vector representation while preserving the intrinsic graph properties. In this review, we first explain the graph embedding task and its challenges. Next, we review a wide range of graph embedding techniques with insights. Then, we evaluate several stat-of-the-art methods against small and large data sets and compare their performance. Finally, potential applications and future directions are presented.

We study regional similarities and differences in language use on an anonymous mobile chat application in the German-speaking area. We use a neural network on 2.3 million online conversations to automatically learn representations of words and cities. These linguistic-use-based representations capture regional distinctions in a high-dimensional vector space that can be clustered and visualized to discover patterns in the data. We find that the resulting regional patterns are closely linked to the traditional division of German dialects, even though most of the conversations are written in standard German. The resulting maps correspond to traditional dialect divisions and language-external spatial structures, with a few notable exceptions that can be explained through external factors.

Our method also facilitates two qualitative analyses, allowing us to discover geographically-pertinent words for various regional levels, as well as creating regional group-specific style profiles based on various linguistic resources. The results of our study strongly suggest the existence of region-specific patterns of language use (“digital regiolects”) representing distinctive strategies of linguistic stylization in relation to linguistic resources and topics. As a methodological contribution, we show how linguistic theory can drive the application and direction of neural network-based representation learning, and how their judicious application provides the basis for qualitative analysis of large-scale data collections.

Current approaches to learning semantic representations of sentences often use prior word-level knowledge. The current study aims to leverage visual information in order to capture sentence level semantics without the need for word embeddings. We use a multimodal sentence encoder trained on a corpus of images with matching text captions to produce visually grounded sentence embeddings. Deep Neural Networks are trained to map the two modalities to a common embedding space such that for an image the corresponding caption can be retrieved and vice versa. We show that our model achieves results comparable to the current state of the art on two popular image-caption retrieval benchmark datasets: Microsoft Common Objects in Context (MSCOCO) and Flickr8k. We evaluate the semantic content of the resulting sentence embeddings using the data from the Semantic Textual Similarity (STS) benchmark task and show that the multimodal embeddings correlate well with human semantic similarity judgements. The system achieves state-of-the-art results on several of these benchmarks, which shows that a system trained solely on multimodal data, without assuming any word representations, is able to capture sentence level semantics. Importantly, this result shows that we do not need prior knowledge of lexical level semantics in order to model sentence level semantics. These findings demonstrate the importance of visual information in semantics.

We present two studies on neural network architectures that learn to represent sentences by composing their words according to automatically induced binary trees, without ever being shown a correct parse tree. We use Tree-Long Short-Term Memories (LSTMs) as our composition function, applied along a tree structure found by a differentiable natural language chart parser. The models simultaneously optimise both the composition function and the parser, thus eliminating the need for externally provided parse trees, which are normally required for Tree-LSTMs. They can therefore be seen as tree-based recurrent neural networks that are unsupervised with respect to the parse trees. Due to being fully differentiable, the models are easily trained with an off-the-shelf gradient descent method and backpropagation.

In the first part of this paper, we introduce a model based on the CKY chart parser, and evaluate its downstream performance on a natural language inference task and a reverse dictionary task. Further, we show how its performance can be improved with an attention mechanism which fully exploits the parse chart, by attending over all possible subspans of the sentence. We find that our approach is competitive against similar models of comparable size and outperforms Tree-LSTMs that use trees produced by a parser.

Finally, we present an alternative architecture based on a shift-reduce parser. We perform an analysis of the trees induced by both our models, to investigate whether they are consistent with each other and across re-runs, and whether they resemble the trees produced by a standard parser.

Sentence-level representations are necessary for various natural language processing tasks. Recurrent neural networks have proven to be very effective in learning distributed representations and can be trained efficiently on natural language inference tasks. We build on top of one such model and propose a hierarchy of bidirectional LSTM and max pooling layers that implements an iterative refinement strategy and yields state of the art results on the SciTail dataset as well as strong results for Stanford Natural Language Inference and Multi-Genre Natural Language Inference. We can show that the sentence embeddings learned in this way can be utilized in a wide variety of transfer learning tasks, outperforming InferSent on 7 out of 10 and SkipThought on 8 out of 9 SentEval sentence embedding evaluation tasks. Furthermore, our model beats the InferSent model in 8 out of 10 recently published SentEval probing tasks designed to evaluate sentence embeddings’ ability to capture some of the important linguistic properties of sentences.