The deployment of deep neural networks on resource-constrained devices relies on quantization. While static, uniform quantization applies a fixed bit-width to all inputs, it fails to adapt to their varying complexity. Dynamic, instance-based mixed-precision quantization promises a superior accuracy-efficiency trade-off by allocating higher precision only when needed. However, a critical bottleneck remains: existing methods require a costly dequantize-to-float and requantize-to-integer cycle to change precision, breaking the integer-only hardware paradigm and compromising performance gains. This paper introduces Dynamic Quantization Training (DQT), a novel framework that removes this bottleneck. At the core of DQT is a nested integer representation where lower-precision values are bit-wise embedded within higher-precision ones. This design, coupled with custom integer-only arithmetic, allows for on-the-fly bit-width switching through a near-zero-cost bit-shift operation. This makes DQT the first quantization framework to enable both dequantization-free static mixed-precision of the backbone network, and truly efficient dynamic, instance-based quantization through a lightweight controller that decides at runtime how to quantize each layer. We demonstrate DQT state-of-the-art performance on ResNet18 on CIFAR-10 and ResNet50 on ImageNet. On ImageNet, our 4-bit dynamic ResNet50 achieves 77.00% top-1 accuracy, an improvement over leading static (LSQ, 76.70%) and dynamic (DQNET, 76.94%) methods at a comparable BitOPs budget. Crucially, DQT achieves this with a bit-width transition cost of only 28.3M simple bit-shift operations, a drastic improvement over the 56.6M costly Multiply-Accumulate (MAC) floating-point operations required by previous dynamic approaches - unlocking a new frontier in efficient, adaptive AI.

Resource-constrained smart eyewear (SEW) devices face significant challenges when deploying deep neural networks due to limited computational capacity and battery life. Computational offloading to companion devices like smartphones and cloud servers addresses processing limitations, but data transmission becomes a critical bottleneck, consuming over 50% of total energy in some scenarios. Although lossless compression methods provide limited data reduction for intermediate tensors, lossy techniques such as Vector Quantization (VQ) offer higher compression ratios (requiring only 3.3 bits per float) at the expense of inference accuracy degradation. This paper presents an adaptive multi-stage compression framework that dynamically balances these trade-offs across the SEW-phone-cloud continuum. We employ VQ at the SEW-phone interface where aggressive compression is essential (achieving 89.6% tensor size reduction with 90% retained accuracy), followed by adaptive selection between quantization and run-length encoding for phone-to-cloud transmission based on network conditions. A Deep Q-Network (DQN) agent jointly optimizes network partitioning points and compression strategies to minimize energy consumption while preserving accuracy and meeting latency constraints. A large simulation campaign considering object detection and human pose estimation tasks demonstrate that our method achieves 55--70% energy savings and 86--91% violation reduction compared to Neurosurgeon (a dynamic partitioning baseline without compression), 45.8% energy savings versus local execution, and 61.1% savings over uncompressed offloading, with latency violation rates below 9% and acceptable accuracy loss (8.0--8.1%). These results enable practical deployment of AI applications on battery-limited SEW devices.

In the field of medical diagnosis and patient monitoring, effective pattern recognition in neurological time-series data is essential. Traditional methods predominantly based on statistical or probabilistic learning and inference often struggle with multivariate, multi-source, state-varying, and noisy data while also posing privacy risks due to excessive information collection and modeling. Furthermore, these methods often overlook critical statistical information, such as the distribution of data points and inherent uncertainties. To address these challenges, we introduce an information theory-based pipeline that leverages specialized features to identify patterns in neurological time-series data while minimizing privacy risks. We incorporate various entropy methods based on the characteristics of different scenarios and entropy. For stochastic state transition applications, we incorporate Shannon's entropy, entropy rates, entropy production, and the von Neumann entropy of Markov chains. When state modeling is impractical, we select and employ approximate entropy, increment entropy, dispersion entropy, phase entropy, and slope entropy. The pipeline's effectiveness and scalability are demonstrated through pattern analysis in a dementia care dataset and also an epileptic and a myocardial infarction dataset. The results indicate that our information theory-based pipeline can achieve average performance improvements across various models on the recall rate, F1 score, and accuracy by up to 13.08 percentage points, while enhancing inference efficiency by reducing the number of model parameters by an average of 3.10 times. Thus, our approach opens a promising avenue for improved, efficient, and critical statistical information-considered pattern recognition in medical time-series data.

This work presents a method for tactile classification of materials for virtual reality (VR) based robot teleoperation. In our system, a human-operator uses a remotely controlled robot-manipulator with an optical fibre-based tactile and proximity sensor to scan surfaces of objects in a remote environment. Tactile and proximity data and the robot's end-effector state feedback are used for the classification of objects' materials which are then visualized in the VR reconstruction of the remote environment for each object. Machine learning techniques such as random forest, convolutional neural and multi-modal convolutional neural networks were used for material classification. The proposed system and methods were tested with five different materials and classification accuracy of 90 % and more was achieved. The results of material classification were successfully exploited for visualising the remote scene in the VR interface to provide more information to the human-operator.

Remote characterisation of the environment during physical robot-environment interaction is an important task commonly accomplished in telerobotics. This paper demonstrates how tactile and proximity sensing can be efficiently used to perform automatic crack detection. A custom-designed integrated tactile and proximity sensor is implemented. It measures the deformation of its body when interacting with the physical environment and distance to the environment's objects with the help of fibre optics. This sensor was used to slide across different surfaces and the data recorded during the experiments was used to detect and classify cracks, bumps and undulations. The proposed method uses machine learning techniques (mean absolute value as feature and random forest as classifier) to detect cracks and determine their width. An average crack detection accuracy of 86.46% and width classification accuracy of 57.30% is achieved. Kruskal-Wallis results (p<; 0.001) indicate statistically significant differences among results obtained when analysing only force data, only proximity data and both force and proximity data. In contrast to previous techniques, which mainly rely on visual modality, the proposed approach based on optical fibres is suitable for operation in extreme environments, such as nuclear facilities in which nuclear radiation may damage the electronic components of video cameras.

Computer vision for biomedical imaging applications is fast developing and at once demanding field of computer science. In particular, computer vision technique provides excellent results for detection and segmentation problems in tomographic imaging. X-ray phase contrast Tomography (XPCT) is a noninvasive 3D imaging technique with high sensitivity for soft tissues. Despite a considerable progress in XPCT data acquisition and data processing methods, the problem in degradation of image quality due to artifacts remains a widespread and often critical issue for computer vision applications. One of the main problems originates from a sample alteration during a long tomographic scan. We proposed and tested Simultaneous Iterative Reconstruction algorithm with Total Variation regularization to reduce the number of projections in high resolution XPCT scans of ex-vivo mouse spinal cord. We have shown that the proposed algorithm allows tenfold reducing the number of projections and, therefore, the exposure time, with conservation of the important morphological information in 3D image with quality acceptable for computer graphics and computer vision applications. Our research paves a way for more effective implementation of advanced computer technologies in phase contrast tomographic research.