Fine-grained image recognition is central to many multimedia tasks such as search, retrieval and captioning. Unfortunately, these tasks are still challenging since the appearance of samples of the same class can be more different than those from different classes. This issue is mainly due to changes in deformation, pose, and the presence of clutter. In the literature, attention has been one of the most successful strategies to handle the aforementioned problems. Attention has been typically implemented in neural networks by selecting the most informative regions of the image that improve classification. In contrast, in this paper, attention is not applied at the image level but to the convolutional feature activations. In essence, with our approach, the neural model learns to attend to lower-level feature activations without requiring part annotations and uses those activations to update and rectify the output likelihood distribution. The proposed mechanism is modular, architecture-independent and efficient in terms of both parameters and computation required. Experiments demonstrate that well-known networks such as Wide Residual Networks and ResNeXt, when augmented with our approach, systematically improve their classification accuracy and become more robust to changes in deformation and pose and to the presence of clutter. As a result, our proposal reaches state-of-the-art classification accuracies in CIFAR-10, the Adience gender recognition task, Stanford Dogs, and UEC-Food100 while obtaining competitive performance in Ima-geNet, CIFAR-100, CUB200 Birds, and Stanford Cars. In addition, we analyze the different components of our model, showing that the proposed attention modules succeed in finding the most discriminative regions of the image. Finally, as a proof of concept, we demonstrate that with only local predictions, an augmented neural network can successfully classify an image before reaching any fully connected layer, thus reducing the computational amount up to 10%.

It can be argued that finding an interpretable lowdimensional representation of a potentially highdimensional phenomenon is central to the scientific enterprise. Independent component analysis (ICA) refers to an ensemble of methods which formalize this goal and provide estimation procedure for practical application. This work proposes mechanism sparsity regularization as a new principle to achieve nonlinear ICA when latent factors depend sparsely on observed auxiliary variables and/or past latent factors. We show that the latent variables can be recovered up to a permutation if one regularizes the latent mechanisms to be sparse and if some graphical criterion is satisfied by the data generating process. As a special case, our framework shows how one can leverage unknown-target interventions on the latent factors to disentangle them, thus drawing further connections between ICA and causality. We validate our theoretical results with toy experiments.

No abstract

Objectives: A substantial share of type 2 diabetes mellitus (T2DM) patients receives insulin. However, little is known about the real-world treatment patterns around insulin initiation. Methods: This was a retrospective claims data analysis. T2DM patients who initiated an insulin therapy between 01/01/2013 and 31/12/2015 were identified in the German AOK PLUS dataset. For validation of results, additional data on a similar T2DM patient population were collected in a Germany-wide medical chart review. Results: 284,878 T2DM patients were identified. 27,340 (9.6%) of these initiated an insulin treatment during the inclusion period (mean age: 72.2 years; 51.4% female). Mean/median weight j BMI of patients with available clinical data was 85.8/84.0 kg (SD:18.9) j 30.6/29.8 kg/m 2 (SD:6.1) at baseline. Mean/median HbA1c-value at baseline was 8.4/8.0% (SD: 1.8). Most commonly prescribed antidiabetic drugs (AD) within 6 months before insulin initiation were metformin (MET; 54.0%), DPP-4 inhibitors (DPP-4i; 37.6%) and sulfonylureas (SU; 29.5%). 23.2% of patients did not receive any AD prescription within 6 months before insulin initiation. 10,953 of above 27,340 insulin starters (40.1%) initiated their insulin therapy without concomitant ADs (insulin monotherapy); 43% of these patients did not receive any AD before insulin initiation. Of the remaining 16,387 patients (59.9%), 4,070 patients (14.9%) received MET only as concomitant AD, 6,385 (23.4%) received MET plus at least one further AD, and 5,932 (21.7%) received at least one further AD excluding MET. Throughout the first year of treatment, prescribed insulin dosage increased over time, resulting in approximately 43.3-77.9 IUs per observed patient day after 12 months of insulin treatment. Conclusions: Characteristics of German T2DM patients initiating insulin deviate substantially from the average German population, especially in terms of weight. We identified an unexpectedly high number of patients without previous AD therapy receiving insulin monotherapy, which is not in line with the clinical guidelines.