Conventional Unsupervised Domain Adaptation (UDA) methods presume source and target domain data to be simultaneously available during training. Such an assumption may not hold in practice, as source data is often inaccessible (e.g., due to privacy reasons). On the contrary, a pre-trained source model is usually available, which performs poorly on target due to the well-known domain shift problem. This translates into a significant amount of misclassifications, which can be interpreted as structured noise affecting the inferred target pseudo-labels. In this work, we cast UDA as a pseudo-label refinery problem in the challenging source-free scenario. We propose Negative Ensemble Learning (NEL) technique, a unified method for adaptive noise filtering and progressive pseudo-label refinement. NEL is devised to tackle noisy pseudo-labels by enhancing diversity in ensemble members with different stochastic (i) input augmentation and (ii) feedback. The latter is achieved by leveraging the novel concept of Disjoint Residual Labels, which allow propagating diverse information to the different members. Eventually, a single model is trained with the refined pseudo-labels, which leads to a robust performance on the target domain. Extensive experiments show that the proposed method achieves state-of-the-art performance on major UDA benchmarks, such as Digit5, PACS, Visda-C, and DomainNet, without using source data samples at all.

Fermentation is a widely used process in the biotechnology industry, in which sugar-based substrates are transformed into a new product through chemical reactions carried out by microorganisms. Fermentation yields depend heavily on critical process parameter (CPP) values which need to be finely tuned throughout the process; this is usually performed by a biotech production expert relying on empirical rules and personal experience. Although developing a mathematical model to analytically describe how yields depend on CPP values is too challenging because the process involves living organisms, we demonstrate the benefits that can be reaped by using a black-box machine learning (ML) approach based on recurrent neural networks (RNN) and long short-term memory (LSTM) neural networks to predict real time OD_600nm values from fermentation CPP time series. We tested both networks on an E. coli fermentation process (upstream) optimized to obtain inclusion bodies whose purification (downstream) in a later stage will yield a targeted neurotrophin recombinant protein. We achieved root mean squared error (RMSE) and relative error on final yield (REFY) performances which demonstrate that RNN and LSTM are indeed promising approaches for real-time, in-line process yield estimation, paving the way for machine learning-based fermentation process control algorithms.

In this paper, we present an approach for estimating the leaf density of trees while navigating in a forest. To this end, we consider an Unmanned Aerial Vehicle (UAV) equipped with a biosonar sensor that mimics the sonar sensors of echolocating bats. Such sensors provide a light-weight and cost-effective alternative to other widely used sensors such as camera, LiDAR and are gaining popularity among the robotics research community. The obtained echo signals during UAV navigation are processed to obtain the leaf density in the main lobe of the sonar first using a mel spectogram and then a Deep Convolutional Neural Network (CNN) trained on a set of known environment. We further evaluate our approach in simulation by considering trees with different leaf density (that is, resolution). It is seen that our method achieves promising results with an accuracy of 98.7%.

Vision loss is often caused by retinal disorders, such as age-related macular degeneration and diabetic retinopathy, where early indicators like microaneurysms and hemorrhages appear as changes in retinal blood vessels. Accurate segmentation of these vessels in retinal images is essential for early diagnosis. However, retinal vessel segmentation presents challenges due to complex vessel structures, low contrast, and dense branching patterns, which are further complicated in resource-limited settings requiring lightweight solutions. To address these challenges, we propose a novel Lightweight Hypercomplex U-Net¹ (LHUN) with Vessel Thickness-Guided Dice Loss (VTDL), collectively called LHU-VT. LHUN utilizes hypercomplex octonions to capture intricate patterns and cross-channel relationships in fundus images, reducing parameter count and enabling edge deployment. The VTDL component applies vessel thickness-guided weights to address class imbalance, thereby enhancing segmentation accuracy. Our experiments show that LHU-VT significantly outperforms current methods, achieving up to 2.4× fewer FLOPs, 4.4× fewer parameters, and 2.6× smaller model size. The model achieves AUC scores of 0.9938, 0.9879, 0.9988, and 0.9808, respectively, on four benchmark datasets CHASE, DRIVE, STARE, and HRF.

The concept of compressing deep Convolutional Neural Networks (CNNs) is essential to use limited computation, power, and memory resources on embedded devices. However, existing methods achieve this objective at the cost of a drop in inference accuracy in computer vision tasks. To address such a drawback, we propose a framework that leverages knowledge distillation along with customizable block-wise optimization to learn a lightweight CNN structure while preserving better control over the compression-performance tradeoff. Considering specific resource constraints, e.g., floating-point operations per inference (FLOPs) or model-parameters, our method results in a state of the art network compression while being capable of achieving better inference accuracy. In a comprehensive evaluation, we demonstrate that our method is effective, robust, and consistent with results over a variety of network architectures and datasets, at negligible training overhead. In particular, for the already compact network MobileNet_v2, our method offers up to 2× and 5.2× better model compression in terms of FLOPs and model-parameters, respectively, while getting 1.05% better model performance than the baseline network.

In this paper, we investigate brain activity associated with complex visual tasks, showing that electroencephalography (EEG) data can help computer vision in reliably recognizing actions from video footage that is used to stimulate human observers. Notably, we consider not only typical “explicit” video action benchmarks, but also more complex data sequences in which action concepts are only referred to, implicitly. To this end, we consider a challenging action recognition benchmark dataset—Moments in Time—whose video sequences do not explicitly visualize actions, but only implicitly refer to them (e.g., fireworks in the sky as an extreme example of “flying”). We employ such videos as stimuli and involve a large sample of subjects to collect a high-definition, multi-modal EEG and video data, designed for understanding action concepts. We discover an agreement among brain activities of different subjects stimulated by the same video footage. We name it as subjects consensus, and we design a computational pipeline to transfer knowledge from EEG to video, sharply boosting the recognition performance.