In natural language understanding (NLU) production systems, users' evolving needs necessitate the addition of new features over time, indexed by new symbols added to the meaning representation space. This requires additional training data and results in ever-growing datasets. We present the first systematic investigation of this incremental symbol learning scenario. Our analysis reveals a troubling quirk in building broad-coverage NLU systems: as the training dataset grows, performance on the new symbol often decreases if we do not accordingly increase its training data. This suggests that it becomes more difficult to learn new symbols with a larger training dataset. We show that this trend holds for multiple mainstream models on two common NLU tasks: intent recognition and semantic parsing. Rejecting class imbalance as the sole culprit, we reveal that the trend is closely associated with an effect we call source signal dilution, where strong lexical cues for the new symbol become diluted as the training dataset grows. Selectively dropping training examples to prevent dilution often reverses the trend, showing the over-reliance of mainstream neural NLU models on simple lexical cues. Code, models, and data are available at https://aka.ms/nlu-incremental-symbol-learning

Mission teams are exposed to the emotional toll of life and death decisions. These are small groups of specially trained people supported by intelligent machines for dealing with stressful environments and scenarios. We developed a composite model for stress monitoring in such teams of human and autonomous machines. This modelling aims to identify the conditions that may contribute to mission failure. The proposed model is composed of three parts: 1) a computational logic part that statically describes the stress states of teammates; 2) a decision part that manifests the mission status at any time; 3) a stress propagation part based on standard Susceptible-Infected-Susceptible (SIS) paradigm. In contrast to the approaches such as agent-based, random-walk and game models, the proposed model combines various mechanisms to satisfy the conditions of stress propagation in small groups. Our core approach involves data structures such as decision tables and decision diagrams. These tools are adaptable to human-machine teaming as well.

Data-driven interatomic potentials have emerged as a powerful class of surrogate models for {\it ab initio} potential energy surfaces that are able to reliably predict macroscopic properties with experimental accuracy. In generating accurate and transferable potentials the most time-consuming and arguably most important task is generating the training set, which still requires significant expert user input. To accelerate this process, this work presents \text{\it hyperactive learning} (HAL), a framework for formulating an accelerated sampling algorithm specifically for the task of training database generation. The key idea is to start from a physically motivated sampler (e.g., molecular dynamics) and add a biasing term that drives the system towards high uncertainty and thus to unseen training configurations. Building on this framework, general protocols for building training databases for alloys and polymers leveraging the HAL framework will be presented. For alloys, ACE potentials for AlSi10 are created by fitting to a minimal HAL-generated database containing 88 configurations (32 atoms each) with fast evaluation times of <100 microsecond/atom/cpu-core. These potentials are demonstrated to predict the melting temperature with excellent accuracy. For polymers, a HAL database is built using ACE, able to determine the density of a long polyethylene glycol (PEG) polymer formed of 200 monomer units with experimental accuracy by only fitting to small isolated PEG polymers with sizes ranging from 2 to 32.

Chain event graphs (CEGs) are a family of probabilistic graphical models that were first proposed in Smith and Anderson (2008) as an alternative to the family of Bayesian networks (BNs). In particular, CEGs were developed to explicitly accommodate processes exhibiting asymmetries of two types: (1) asymmetric independence structures or context-specific conditional independences where some statistical independences hold for certain values of the conditioning variables but not the others; and (2) asymmetric event spaces which are precisely event spaces that do not admit a product space structure. The latter asymmetry arises due to the presence of structural zeros and structural missing values, often-times by design (Shenvi & Smith, 2020). For example, consider modelling hospitalisations arising from infection caused by a circulating virus, and suppose that one of the two strains (call it strain A) of the virus has no treatment currently available while the other has a choice of two possible treatments. On the one hand, a variable of "Treatment" with state space {Treatment 1, Treatment 2} would be structurally missing and have no sensible value for those infected by strain A of the virus. Whereas on the other hand, if its state space is redefined to be {Treatment 1, Treatment 2, No treatment} then Treatment 1 and Treatment 2 would have structurally zero counts for those infected by strain A, i.e. irrespective of the sample size, there would always be zero individuals who are treated with either Treatment 1 or Treatment 2 among those infected by strain A. Such a process is inherently asymmetric. BNs, being variable-based - i.e. they use variables as the building blocks of their models - are unable to fully describe such asymmetries within their underlying statistical model and graphical structure.

Causal learning has attracted much attention in recent years because causality reveals the essential relationship between things and indicates how the world progresses. However, there are many problems and bottlenecks in traditional causal learning methods, such as high-dimensional unstructured variables, combinatorial optimization problems, unknown intervention, unobserved confounders, selection bias and estimation bias. Deep causal learning, that is, causal learning based on deep neural networks, brings new insights for addressing these problems. While many deep learning-based causal discovery and causal inference methods have been proposed, there is a lack of reviews exploring the internal mechanism of deep learning to improve causal learning. In this article, we comprehensively review how deep learning can contribute to causal learning by addressing conventional challenges from three aspects: representation, discovery, and inference. We point out that deep causal learning is important for the theoretical extension and application expansion of causal science and is also an indispensable part of general artificial intelligence. We conclude the article with a summary of open issues and potential directions for future work.

Transfer Learning is an area of statistics and machine learning research that seeks answers to the following question: how do we build successful learning algorithms when the data available for training our model is qualitatively different from the data we hope the model will perform well on? In this thesis, we focus on a specific area of Transfer Learning called label shift, also known as quantification. In quantification, the aforementioned discrepancy is isolated to a shift in the distribution of the response variable. In such a setting, accurately inferring the response variable's new distribution is both an important estimation task in its own right and a crucial step for ensuring that the learning algorithm can adapt to the new data. We make two contributions to this field. First, we present a new procedure called SELSE which estimates the shift in the response variable's distribution. Second, we prove that SELSE is semiparametric efficient among a large family of quantification algorithms, i.e., SELSE's normalized error has the smallest possible asymptotic variance matrix compared to any other algorithm in that family. This family includes nearly all existing algorithms, including ACC/PACC quantifiers and maximum likelihood based quantifiers such as EMQ and MLLS. Empirical experiments reveal that SELSE is competitive with, and in many cases outperforms, existing state-of-the-art quantification methods, and that this improvement is especially large when the number of test samples is far greater than the number of train samples.

The Bayesian approach to solving inverse problems relies on the choice of a prior. This critical ingredient allows the formulation of expert knowledge or physical constraints in a probabilistic fashion and plays an important role for the success of the inference. Recently, Bayesian inverse problems were solved using generative models as highly informative priors. Generative models are a popular tool in machine learning to generate data whose properties closely resemble those of a given database. Typically, the generated distribution of data is embedded in a low-dimensional manifold. For the inverse problem, a generative model is trained on a database that reflects the properties of the sought solution, such as typical structures of the tissue in the human brain in magnetic resonance (MR) imaging. The inference is carried out in the low-dimensional manifold determined by the generative model which strongly reduces the dimensionality of the inverse problem. However, this proceeding produces a posterior that admits no Lebesgue density in the actual variables and the accuracy reached can strongly depend on the quality of the generative model. For linear Gaussian models we explore an alternative Bayesian inference based on probabilistic generative models which is carried out in the original high-dimensional space. A Laplace approximation is employed to analytically derive the required prior probability density function induced by the generative model. Properties of the resulting inference are investigated. Specifically, we show that derived Bayes estimates are consistent, in contrast to the approach employing the low-dimensional manifold of the generative model. The MNIST data set is used to construct numerical experiments which confirm our theoretical findings.

Humans demonstrate a variety of interesting behavioral characteristics when performing tasks, such as selecting between seemingly equivalent optimal actions, performing recovery actions when deviating from the optimal trajectory, or moderating actions in response to sensed risks. However, imitation learning, which attempts to teach robots to perform these same tasks from observations of human demonstrations, often fails to capture such behavior. Specifically, commonly used learning algorithms embody inherent contradictions between the learning assumptions (e.g., single optimal action) and actual human behavior (e.g., multiple optimal actions), thereby limiting robot generalizability, applicability, and demonstration feasibility. To address this, this paper proposes designing imitation learning algorithms with a focus on utilizing human behavioral characteristics, thereby embodying principles for capturing and exploiting actual demonstrator behavioral characteristics. This paper presents the first imitation learning framework, Bayesian Disturbance Injection (BDI), that typifies human behavioral characteristics by incorporating model flexibility, robustification, and risk sensitivity. Bayesian inference is used to learn flexible non-parametric multi-action policies, while simultaneously robustifying policies by injecting risk-sensitive disturbances to induce human recovery action and ensuring demonstration feasibility. Our method is evaluated through risk-sensitive simulations and real-robot experiments (e.g., table-sweep task, shaft-reach task and shaft-insertion task) using the UR5e 6-DOF robotic arm, to demonstrate the improved characterisation of behavior. Results show significant improvement in task performance, through improved flexibility, robustness as well as demonstration feasibility.

Although numerous algorithms have been proposed to solve the categorical data clustering problem, how to access the statistical significance of a set of categorical clusters remains unaddressed. To fulfill this void, we employ the likelihood ratio test to derive a test statistic that can serve as a significance-based objective function in categorical data clustering. Consequently, a new clustering algorithm is proposed in which the significance-based objective function is optimized via a Monte Carlo search procedure. As a by-product, we can further calculate an empirical $p$-value to assess the statistical significance of a set of clusters and develop an improved gap statistic for estimating the cluster number. Extensive experimental studies suggest that our method is able to achieve comparable performance to state-of-the-art categorical data clustering algorithms. Moreover, the effectiveness of such a significance-based formulation on statistical cluster validation and cluster number estimation is demonstrated through comprehensive empirical results.

Cross entropy loss has served as the main objective function for classification-based tasks. Widely deployed for learning neural network classifiers, it shows both effectiveness and a probabilistic interpretation. Recently, after the success of self supervised contrastive representation learning methods, supervised contrastive methods have been proposed to learn representations and have shown superior and more robust performance, compared to solely training with cross entropy loss. However, cross entropy loss is still needed to train the final classification layer. In this work, we investigate the possibility of learning both the representation and the classifier using one objective function that combines the robustness of contrastive learning and the probabilistic interpretation of cross entropy loss. First, we revisit a previously proposed contrastive-based objective function that approximates cross entropy loss and present a simple extension to learn the classifier jointly. Second, we propose a new version of the supervised contrastive training that learns jointly the parameters of the classifier and the backbone of the network. We empirically show that our proposed objective functions show a significant improvement over the standard cross entropy loss with more training stability and robustness in various challenging settings.