Machine learning typically presupposes classical probability theory which implies that aggregation is built upon expectation. There are now multiple reasons to motivate looking at richer alternatives to classical probability theory as a mathematical foundation for machine learning. We systematically examine a powerful and rich class of alternative aggregation functionals, known variously as spectral risk measures, Choquet integrals or Lorentz norms. We present a range of characterization results, and demonstrate what makes this spectral family so special. In doing so we arrive at a natural stratification of all coherent risk measures in terms of the upper probabilities that they induce by exploiting results from the theory of rearrangement invariant Banach spaces. We empirically demonstrate how this new approach to uncertainty helps tackling practical machine learning problems.

Deep neural networks (DNNs) have become increasingly important due to their excellent empirical performance on a wide range of problems. However, regularization is generally achieved by indirect means, largely due to the complex set of functions defined by a network and the difficulty in measuring function complexity. There exists no method in the literature for additive regularization based on a norm of the function, as is classically considered in statistical learning theory. In this work, we propose sampling-based approximations to weighted function norms as regularizers for deep neural networks. We provide, to the best of our knowledge, the first proof in the literature of the NP-hardness of computing function norms of DNNs, motivating the necessity of a stochastic optimization strategy. Based on our proposed regularization scheme, stability-based bounds yield a $\mathcal{O}(N^{-\frac{1}{2}})$ generalization error for our proposed regularizer when applied to convex function sets. We demonstrate broad conditions for the convergence of stochastic gradient descent on our objective, including for non-convex function sets such as those defined by DNNs. Finally, we empirically validate the improved performance of the proposed regularization strategy for both convex function sets as well as DNNs on real-world classification and segmentation tasks.

Deep neural networks have had an enormous impact on image analysis. State-of-the-art training methods, based on weight decay and DropOut, result in impressive performance when a very large training set is available. However, they tend to have large problems overfitting to small data sets. Indeed, the available regularization methods deal with the complexity of the network function only indirectly. In this paper, we study the feasibility of directly using the $L_2$ function norm for regularization. Two methods to integrate this new regularization in the stochastic backpropagation are proposed. Moreover, the convergence of these new algorithms is studied. We finally show that they outperform the state-of-the-art methods in the low sample regime on benchmark datasets (MNIST and CIFAR10). The obtained results demonstrate very clear improvement, especially in the context of small sample regimes with data laying in a low dimensional manifold. Source code of the method can be found at \url{https://github.com/AmalRT/DNN_Reg}.

Natural Language Processing (NLP) tasks, such as Named Entity Recognition (NER), involve an iterative process of model optimization to identify different types of words or semantic entities. This optimization to achieve a more precise model becomes computationally difficult as the number of iterations increase. The small datasets available for training typically limit the models. Adding iterations on such sets to further optimize the model can often cause over-fitting, which generally leads to reduced performance. Therefore, the choice of convergence criteria is a critical step in robust and accurate model building. We evaluate different convergence criteria in terms of their robustness, stopping threshold selection, and independence from the training data size and entity. The underlying framework employs a limited-memory Broyden-Fletcher-Goldfarb-Shanno (L-BFGS) parameter optimization in the context of Conditional Random Fields (CRF). This paper presents a convergence criterion for robust training irrespective of semantic types and data sizes with two-orders of magnitude reduction in stopping threshold for improved model accuracy and faster convergence. Additionally, we examine convergence with active learning to further reduce the training data and training time.