Abstract:Recently, a new Continual Learning (CL) paradigm was presented to control catastrophic forgetting, called Interval Continual Learning (InterContiNet), which relies on enforcing interval constraints on the neural network parameter space. Unfortunately, InterContiNet training is challenging due to the high dimensionality of the weight space, making intervals difficult to manage. To address this issue, we introduce HyperInterval, a technique that employs interval arithmetic within the embedding space and utilizes a hypernetwork to map these intervals to the target network parameter space. We train interval embeddings for consecutive tasks and train a hypernetwork to transform these embeddings into weights of the target network. An embedding for a given task is trained along with the hypernetwork, preserving the response of the target network for the previous task embeddings. Interval arithmetic works with a more manageable, lower-dimensional embedding space rather than directly preparing intervals in a high-dimensional weight space. Our model allows faster and more efficient training. Furthermore, HyperInterval maintains the guarantee of not forgetting. At the end of training, we can choose one universal embedding to produce a single network dedicated to all tasks. In such a framework, hypernetwork is used only for training and can be seen as a meta-trainer. HyperInterval obtains significantly better results than InterContiNet and gives SOTA results on several benchmarks.
Abstract:Artificial neural networks suffer from catastrophic forgetting when they are sequentially trained on multiple tasks. To overcome this problem, there exist many continual learning strategies. One of the most effective is the hypernetwork-based approach. The hypernetwork generates the weights of a target model based on the task's identity. The model's main limitation is that hypernetwork can produce completely different nests for each task. Consequently, each task is solved separately. The model does not use information from the network dedicated to previous tasks and practically produces new architectures when it learns the subsequent tasks. To solve such a problem, we use the lottery ticket hypothesis, which postulates the existence of sparse subnetworks, named winning tickets, that preserve the performance of a full network. In the paper, we propose a method called HyperMask, which trains a single network for all tasks. Hypernetwork produces semi-binary masks to obtain target subnetworks dedicated to new tasks. This solution inherits the ability of the hypernetwork to adapt to new tasks with minimal forgetting. Moreover, due to the lottery ticket hypothesis, we can use a single network with weighted subnets dedicated to each task.
Abstract:Neural networks, autoencoders in particular, are one of the most promising solutions for unmixing hyperspectral data, i.e. reconstructing the spectra of observed substances (endmembers) and their relative mixing fractions (abundances). Unmixing is needed for effective hyperspectral analysis and classification. However, as we show in this paper, the training of autoencoders for unmixing is highly dependent on weights initialisation. Some sets of weights lead to degenerate or low performance solutions, introducing negative bias in expected performance. In this work we present the results of experiments investigating autoencoders' stability, verifying the dependence of reconstruction error on initial weights and exploring conditions needed for successful optimisation of autoencoder parameters.
Abstract:In recent years, a lot of attention is paid to deep learning methods in the context of vision-based construction site safety systems, especially regarding personal protective equipment. However, despite all this attention, there is still no reliable way to establish the relationship between workers and their hard hats. To answer this problem a combination of deep learning, object detection and head keypoint localization, with simple rule-based reasoning is proposed in this article. In tests, this solution surpassed the previous methods based on the relative bounding box position of different instances, as well as direct detection of hard hat wearers and non-wearers. The results show that the conjunction of novel deep learning methods with humanly-interpretable rule-based systems can result in a solution that is both reliable and can successfully mimic manual, on-site supervision. This work is the next step in the development of fully autonomous construction site safety systems and shows that there is still room for improvement in this area.
Abstract:This study is focused on applying genetic algorithms (GA) to model and band selection in hyperspectral image classification. We use a forensic-inspired data set of seven hyperspectral images with blood and five visually similar substances to test GA-optimised classifiers in two scenarios: when the training and test data come from the same image and when they come from different images, which is a more challenging task due to significant spectra differences. In our experiments we compare GA with a classic model optimisation through grid search. Our results show that GA-based model optimisation can reduce the number of bands and create an accurate classifier that outperforms the GS-based reference models, provided that during model optimisation it has access to examples similar to test data. We illustrate this with experiment highlighting the importance of a validation set.