Feature selection is a critical step in data-driven applications, reducing input dimensionality to enhance learning accuracy, computational efficiency, and interpretability. Existing state-of-the-art methods often require post-selection retraining and extensive hyperparameter tuning, complicating their adoption. We introduce a novel, non-intrusive feature selection layer that, given a target feature count $k$, automatically identifies and selects the $k$ most informative features during neural network training. Our method is uniquely simple, requiring no alterations to the loss function, network architecture, or post-selection retraining. The layer is mathematically elegant and can be fully described by: \begin{align} \nonumber \tilde{x}_i = a_i x_i + (1-a_i)z_i \end{align} where $x_i$ is the input feature, $\tilde{x}_i$ the output, $z_i$ a Gaussian noise, and $a_i$ trainable gain such that $\sum_i{a_i^2}=k$. This formulation induces an automatic clustering effect, driving $k$ of the $a_i$ gains to $1$ (selecting informative features) and the rest to $0$ (discarding redundant ones) via weighted noise distortion and gain normalization. Despite its extreme simplicity, our method achieves competitive performance on standard benchmark datasets and a novel real-world dataset, often matching or exceeding existing approaches without requiring hyperparameter search for $k$ or retraining. Theoretical analysis in the context of linear regression further validates its efficacy. Our work demonstrates that simplicity and performance are not mutually exclusive, offering a powerful yet straightforward tool for feature selection in machine learning.

Recent advancements in Neural Architecture Search (NAS) have introduced methods capable of identifying optimal neural network architectures in minutes on Graphical Processing Units (GPUs) using zero-shot proxies, but mainly focus on single-objective optimization. NAS is also used to discover efficient architectures for edge devices. However, addressing the diverse hardware and application constraints specific to edge platforms remains a significant challenge. In this paper, we introduce a zero-shot NAS approach designed to generate hardware-aware architectures, combined with a selection technique that allows adaptable model optimization across various deployment scenarios without re-executing the search. We demonstrate the flexibility of our solution by benchmarking the Google Coral Edge Tensor Processing Unit (TPU). Our technique led to the efficient exploration of the architecture space of NAS-Bench-201 (NB201) in under a minute, accelerating the search by 25× compared to previous work while maintaining comparable accuracies of 93.24% on CIFAR-10 and 42.99% on ImageNet16-120.

The training phase of deep neural networks requires substantial resources and as such is often performed on cloud servers. However, this raises privacy concerns when the training dataset contains sensitive content, e.g., facial or medical images. In this work, we propose a method to perform the training phase of a deep learning model on both an edge device and a cloud server that prevents sensitive content being transmitted to the cloud while retaining the desired information. The proposed privacy-preserving method uses adversarial early exits to suppress the sensitive content at the edge and transmits the task-relevant information to the cloud. This approach incorporates noise addition during the training phase to provide a differential privacy guarantee. We extensively test our method on different facial and medical datasets with diverse attributes using various deep learning architectures, showcasing its outstanding performance. We also demonstrate the effectiveness of privacy preservation through successful defenses against different white-box, deep and GAN-based reconstruction attacks. This approach is designed for resource-constrained edge devices, ensuring minimal memory usage and computational overhead.

Deep neural networks (DNNs) provide state-of-the-art accuracy for vision tasks, but they require significant resources for training. Thus, they are trained on cloud servers far from the edge devices that acquire the data. This issue increases communication cost, runtime, and privacy concerns. In this study, a novel hierarchical training method for DNNs is proposed that uses early exits in a divided architecture between edge and cloud workers to reduce the communication cost, training runtime, and privacy concerns. The method proposes a brand-new use case for early exits to separate the backward pass of neural networks between the edge and the cloud during the training phase. We address the issues of most available methods that, due to the sequential nature of the training phase, cannot train the levels of hierarchy simultaneously or they do it with the cost of compromising privacy. In contrast, our method can use both edge and cloud workers simultaneously, does not share the raw input data with the cloud, and does not require communication during the backward pass. Several simulations and on-device experiments for different neural network architectures demonstrate the effectiveness of this method. It is shown that the proposed method reduces the training runtime for VGG-16 and ResNet-18 architectures by 29% and 61% in CIFAR-10 classification and by 25% and 81% in Tiny ImageNet classification, respectively, when the communication with the cloud is done over a low bit rate channel. This gain in the runtime is achieved, while the accuracy drop is negligible. This method is advantageous for online learning of high-accuracy DNNs on sensor-holding low-resource devices such as mobile phones or robots as a part of an edge-cloud system, making them more flexible in facing new tasks and classes of data.

Learning probabilistic models that can estimate the density of a given set of samples, and generate samples from that density, is one of the fundamental challenges in unsupervised machine learning. We introduce a new generative model based on denoising density estimators (DDEs), which are scalar functions parametrized by neural networks, that are efficiently trained to represent kernel density estimators of the data. Leveraging DDEs, our main contribution is a novel technique to obtain generative models by minimizing the Kullback-Leibler (KL)-divergence directly. We prove that our algorithm for obtaining generative models is guaranteed to converge consistently to the correct solution. Our approach does not require specific network architecture as in normalizing flows (NFs), nor use ordinary differential equation (ODE) solvers as in continuous NFs. Experimental results demonstrate substantial improvement in density estimation and competitive performance in generative model training.

Preserving privacy is a growing concern in our society where cameras are ubiquitous. In this work, we propose a trainable image acquisition method that removes the sensitive information in the optical domain before it reaches the image sensor. The method benefits from a trainable optical convolution kernel, which transmits the desired information whilst filtering out the sensitive information, making it irretrievable against different privacy attacks in the digital domain. This is in contrast with the current digital privacy-preserving methods that are all vulnerable to direct access attacks. Also, in contrast with most of the previous optical privacy-preserving methods that cannot be trained, our method is data-driven and optimized for the specific application at hand. Moreover, there is no additional computation or power burden on the acquisition system since it works passively in the optical domain and can be even used in conjunction with other privacy-preserving techniques in the digital domain. We demonstrate our new, generic method in several scenarios such as smile or open-mouth detection as the desired attribute while the gender or wearing make-up is filtered out as the sensitive content. Through several experiments, we show that this method is able to reduce around$\mathbf {65}\%$of sensitive content while causing a negligible reduction in the desired information. Moreover, we tested our method by deep reconstruction attack and confirmed the ineffectiveness of this attack to reconstruct the original sensitive content. This new method has different use cases such as feedback systems for smart TV content or outdoor advertising.

Reducing energy consumption is a critical point for neural network models running on edge devices. In this regard, reducing the number of multiply-accumulate (MAC) operations of Deep Neural Networks (DNNs) running on edge hardware accelerators will reduce the energy consumption during inference. Spiking Neural Networks (SNNs) are an example of bio-inspired techniques that can further save energy by using binary activations, and avoid consuming energy when not spiking. The networks can be configured for equivalent accuracy on a task through DNN-to-SNN conversion frameworks but their conversion is based on rate coding therefore the synaptic operations can be high. In this work, we look into different techniques to enforce sparsity on the neural network activation maps and compare the effect of different training regularizers on the efficiency of the optimized DNNs and SNNs.

The problem of estimating a surface shape from its observed reflectance properties still remains a challenging task in computer vision. The presence of global illumination effects such as inter-reflections or cast shadows makes the task particularly difficult for non-convex real-world surfaces. State-of-the-art methods for calibrated photometric stereo address these issues using convolutional neural networks (CNNs) that primarily aim to capture either the spatial context among adjacent pixels or the photometric one formed by illuminating a sample from adjacent directions.In this paper, we bridge these two objectives and introduce an efficient fully-convolutional architecture that can leverage both spatial and photometric context simultaneously. In contrast to existing approaches that rely on standard 2D CNNs and regress directly to surface normals, we argue that using separable 4D convolutions and regressing to 2D Gaussian heat-maps severely reduces the size of the network and leads to more stable predictions.Our experimental results on a real-world photometric stereo benchmark show that the proposed approach outperforms the existing published methods in accuracy. The source code for our method is available at https://github.com/DawyD/UNet-PS-4D.

Abstract Thanks to recent advancements in image processing and deep learning techniques, visual surface inspection in production lines has become an automated process as long as all the defects are visible in a single or a few images. However, it is often necessary to inspect parts under many different illumination conditions to capture all the defects. Training deep networks to perform this task requires large quantities of annotated data, which are rarely available and cumbersome to obtain. To alleviate this problem, we devised an original augmentation approach that, given a small image collection, generates rotated versions of the images while preserving illumination effects, something that random rotations cannot do. We introduce three real multi-illumination datasets, on which we demonstrate the effectiveness of our illumination preserving rotation approach. Training deep neural architectures with our approach delivers a performance increase of up to 51% in terms of AuPRC score over using standard rotations to perform data augmentation.

Despite the substantial progress made in deep learning in recent years, advanced approaches remain computationally intensive. The trade-off between accuracy and computation time and energy limits their use in real-time applications on low power and other resource-constrained systems. In this paper, we tackle this fundamental challenge by introducing a hybrid optical-digital implementation of a convolutional neural network (CNN) based on engineering of the point spread function (PSF) of an optical imaging system. This is done by coding an imaging aperture such that its PSF replicates a large convolution kernel of the first layer of a pre-trained CNN. As the convolution takes place in the optical domain, it has zero cost in terms of energy consumption and has zero latency independent of the kernel size. Experimental results on two datasets demonstrate that our approach yields more than two orders of magnitude reduction in the computational cost while achieving near-state-of-the-art accuracy, or equivalently, better accuracy at the same computational cost.