Semi-supervised domain adaptation aims to transfer knowledge from a labeled source domain to a scarcely labeled target domain, despite distribution shifts. The challenge becomes greater when source and target data differ in acquisition modality, as in remote sensing where variations in sensor type (e.g., optical vs. radar), spectral properties (e.g., RGB vs. multispectral), or spatial resolution are common. This challenging scenario, known as Semi-Supervised Heterogeneous Domain Adaptation (SSHDA), requires learning across modalities with limited target labels. In this work, we propose SAHARA (Semi-supervised Adaptation in Heterogeneous domains via conditional Adversarial Representation disentanglement and Adaptive pseudo-labeling), a new method for SSHDA that combines conditional adversarial feature adaptation with dynamic pseudo-labeling to learn domain-invariant features and handle extremely scarce target annotations. Experiments on two heterogeneous remote sensing benchmarks for scene classification, conducted with both convolutional and transformer-based backbones, demonstrate that SAHARA consistently outperforms existing SSHDA and semi-supervised methods. The code is available at https: //TO-BE-DISCLOSED-UPON-ACCEPTANCE.

This paper analyzes the accuracy on the detection of rapeseed fields using Sentinel-1 (S1) time series. Random Forest (RF) and three deep learning (DL) algorithms namely Long Short-Term Memory Fully Convolutional Network (LSTM-FCN), InceptionTime, and Multi-layer Perceptron (MLP) were tested in this study. All four algorithms were used to classify the S1 time series with a large number of ground samples. To test the transferability of classification models, the algorithms were trained on a given year, and then tested on different years. The results demonstrated the high performance of all four algorithms in mapping rapeseed fields when using different years in training and testing phases (F1 between 85.5% and 92.7%, kappa between 0.85 and 0.93).

With the advent of the Sentinel-2 mission and its high revisit frequency, high-resolution time series of optical images, the use of satellite image time series for automatic land cover mapping has fostered. However, one of the main limitations related to this kind of imagery is the presence of clouds, which often hinders its descriptive potential by reducing the actual temporal resolution. Although some common practices exist to enable their use in land cover processing chains, the majority of them aims at reconstructing the time series upstream to the classification task, hence introducing a heavy, error-prone pre-processing step. With the aim of exploiting the capacity of deep learning networks to adaptively combine tasks, in this preliminary study we propose an end-to-end framework that simultaneously perform cloud removal and classification of a Sentinel-2 image time series for the downstream task of land cover mapping. The proposed framework is evaluated over an agricultural area in Northern Benin. Our first results show comparable performances with respect to using state-of-the-art gap filling pre-processing on Sentinel-2 time series, hence motivating further exploration.

Fine-grained classification often requires recognizing specific object parts, such as beak shape and wing patterns for birds. Encouraging a fine-grained classification model to first detect such parts and then using them to infer the class could help us gauge whether the model is indeed looking at the right details better than with interpretability methods that provide a single attribution map. We propose PDiscoNet to discover object parts by using only image-level class labels along with priors encouraging the parts to be: discriminative, compact, distinct from each other, equivariant to rigid transforms, and active in at least some of the images. In addition to using the appropriate losses to encode these priors, we propose to use part-dropout, where full part feature vectors are dropped at once to prevent a single part from dominating in the classification, and part feature vector modulation, which makes the information coming from each part distinct from the perspective of the classifier. Our results on CUB, CelebA, and PartImageNet show that the proposed method provides substantially better part discovery performance than previous methods while not requiring any additional hyper-parameter tuning and without penalizing the classification performance. The code is available at https://github.com/robertdvdk/part_detection

The Gap safe screening technique is a powerful tool to accelerate the convergence of sparse optimization solvers. Its performance is largely based on the ability to determine the smallest “sphere”, centered at a given feasible dual point, that contains the dual solution. This can be achieved through an inner sphere refinement loop, applied at each screening step. In this work, we show that this refinement loop actually converges to the solution of a fixed-point equation for which we derive a closed-form expression for two common loss functions. This allows us to develop an analytic (i.e., non iterative), more concise and theoretically-grounded variant of the sphere refinement step.

Numerical validation is at the core of machine learning research as it allows us to assess the actual impact of new methods, and to confirm the agreement between theory and practice. Yet, the rapid development of the field poses several challenges: researchers are confronted with a profusion of methods to compare, limited transparency and consensus on best practices, as well as tedious re-implementation work. As a result, validation is often very partial, which can lead to wrong conclusions that slow down the progress of research. We propose Benchopt, a collaborative framework to automatize, publish and reproduce optimization benchmarks in machine learning across programming languages and hardware architectures. Benchopt simplifies benchmarking for the community by providing an off-the-shelf tool for running, sharing and extending experiments. To demonstrate its broad usability, we showcase benchmarks on three standard ML tasks: $\ell_2$-regularized logistic regression, Lasso and ResNet18 training for image classification. These benchmarks highlight key practical findings that give a more nuanced view of state-of-the-art for these problems, showing that for practical evaluation, the devil is in the details.

Safe screening rules are powerful tools to accelerate iterative solvers in sparse regression problems. They allow early identification of inactive coordinates (i.e., those not belonging to the support of the solution) which can thus be screened out in the course of iterations. In this paper, we extend the GAP Safe screening rule to the ℓ1-regularized Kullback-Leibler divergence which does not fulfil the regularity assumptions made in previous works. The proposed approach is experimentally validated on synthetic and real count data sets.

Sparse optimization problems are ubiquitous in many fields such as statistics, signal/image processing and machine learning. This has led to the birth of many iterative algorithms to solve them. A powerful strategy to boost the performance of these algorithms is known as safe screening: it allows the early identification of zero coordinates in the solution, which can then be eliminated to reduce the problem's size and accelerate convergence. In this work, we extend the existing Gap Safe screening framework by relaxing the global strong-concavity assumption on the dual cost function. Instead, we exploit local regularity properties, that is, strong concavity on well-chosen subsets of the domain. The non-negativity constraint is also integrated to the existing framework. Besides making safe screening possible to a broader class of functions that includes $\beta$-divergences (e.g., the Kullback-Leibler divergence), the proposed approach also improves upon the existing Gap Safe screening rules on previously applicable cases (e.g., logistic regression). The proposed general framework is exemplified by some notable particular cases: logistic function, $\beta=1.5$ and Kullback-Leibler divergences. Finally, we showcase the effectiveness of the proposed screening rules with different solvers (coordinate descent, multiplicative-update and proximal gradient algorithms) and different datasets (binary classification, hyperspectral and count data).

Accelerating the solution of the Lasso problem becomes crucial when scaling to very high dimensional data. In this paper, we propose a way to combine two existing acceleration techniques: safe screening tests, which simplify the problem by eliminating useless dictionary atoms; and the use of structured dictionaries which are faster to operate with. A structured approximation of the true dictionary is used at the initial stage of the optimization, and we show how to define screening tests which are still safe despite the approximation error. In particular, we extend a state-of-the-art screening test, the GAP SAFE sphere test, to this new setting. The practical interest of the proposed methodology is demonstrated by considerable reductions in simulation time.

The choice of an appropriate frame, or dictionary, is a crucial step in the sparse representation of a given class of signals. Traditional dictionary learning techniques generally lead to unstructured dictionaries that are costly to deploy and train, and do not scale well to higher dimensional signals. In order to overcome such limitation, we propose a learning algorithm that constrains the dictionary to be a sum of Kronecker products of smaller subdictionaries. This approach, named sum of Kronecker products, is demonstrated experimentally in an image denoising application.