Mixed-precision quantization (MPQ) is crucial for deploying deep neural networks on resource-constrained devices, but finding the optimal bit-width for each layer represents a complex combinatorial optimization problem. Current state-of-the-art methods rely on computationally expensive search algorithms or local sensitivity heuristic proxies like the Hessian, which fail to capture the cascading global effects of quantization error. In this work, we argue that the quantization sensitivity of a layer should not be measured by its local properties, but by its impact on the information flow throughout the entire network. We introduce InfoQ, a novel framework for mixed-precision quantization that is training-free in the bit-width search phase. InfoQ assesses layer importance by performing a single forward pass to measure the change in mutual information in the remaining part of the network, thus creating a global sensitivity score. This approach directly quantifies how quantizing one layer degrades the information characteristics of subsequent layers. The resulting scores are used to formulate bit-width allocation as an integer linear programming problem, which is solved efficiently to minimize total sensitivity under a given budget (e.g., model size or BitOps). Our retraining-free search phase provides a superior search-time/accuracy trade-off (using two orders of magnitude less data compared to state-of-the-art methods such as LIMPQ), while yielding up to a 1% accuracy improvement for MobileNetV2 and ResNet18 on ImageNet at high compression rates (14.00x and 10.66x).

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.

Wearable devices running Human Activity Recognition (HAR) on Inertial Measurement Units (IMUs) waste energy by performing continuous classification for each window, even during long periods of unchanged activity. We address this with a lightweight change-detection gate: a non-parametric algorithm based on dynamic template matching that runs continuously at only approximately 16 kFLOPs per step, requires no offline training, and does not need prior definition of target activity classes. The gate invokes the full HAR network only when it detects an activity change, reducing the computational load by over 67% in realistic monitoring settings. The algorithm is evaluated on smart glasses, smartwatch, and smartphone data, requiring only a brief device-specific calibration phase. The gate achieves 98% sensitivity on UCA-EHAR, ensuring no genuine activity transition is missed, while 75 % specificity keeps unnecessary HAR invocations low. Results on WISDM are 97% sensitivity and 76% specificity, demonstrating robustness and flexibility to various settings.

Transformer architectures based on the attention mechanism have revolutionized natural language processing (NLP), driving major breakthroughs across virtually every NLP task. However, their substantial memory and computational requirements still hinder deployment on ultra-constrained devices such as wearables and Internet-of-Things (IoT) units, where available memory is limited to just a few megabytes. To address this challenge, we introduce EmbBERT, a tiny language model (TLM) architecturally designed for extreme efficiency. The model integrates a compact embedding layer, streamlined feed-forward blocks, and an efficient attention mechanism that together enable optimal performance under strict memory budgets. Through this redesign for the extreme edge, we demonstrate that highly simplified transformer architectures remain remarkably effective under tight resource constraints. EmbBERT requires only 2 MB of total memory, and achieves accuracy performance comparable to the ones of state-of-the-art (SotA) models that require a 10 × memory budget. Extensive experiments on the curated TinyNLP benchmark and the GLUE suite confirm that EmbBERT achieves competitive accuracy, comparable to that of larger SotA models, and consistently outperforms downsized versions of BERT and MAMBA of similar size. Furthermore, we demonstrate the model's resilience to 8-bit quantization, which further reduces memory usage to just 781 kB, and the scalability of the EmbBERT architecture across the sub-megabyte to tens-of-megabytes range. Finally, we perform an ablation study demonstrating the positive contributions of all components and the pre-training procedure. All code, scripts, and checkpoints are publicly released to ensure reproducibility: https://github.com/RiccardoBravin/tiny-LLM.

Tiny Machine Learning (TinyML) is a branch of Machine Learning (ML) that constitutes a bridge between the ML world and the embedded system ecosystem (i.e., Internet-of-Things devices, embedded devices, and edge computing units), enabling the execution of ML algorithms on devices constrained in terms of memory, computational capabilities, and power consumption. Video Streaming Analysis (VSA), one of the most interesting tasks of TinyML, consists in scanning a sequence of frames in a streaming manner, with the goal of identifying interesting patterns. Given the strict constraints of these tiny devices, all the current solutions rely on performing a frame-by-frame analysis, hence not exploiting the temporal component in the stream of data. In this paper, we present StreamTinyNet, the first TinyML architecture to perform multiple-frame VSA, enabling a variety of use cases that requires spatial-temporal analysis that were previously impossible to be carried out at a TinyML level. Experimental results on public-available datasets show the effectiveness and efficiency of the proposed solution. Finally, StreamTinyNet has been ported and tested on the Arduino Nicla Vision, showing the feasibility of what proposed.