Deep neural networks are renowned for their ability to generalise well across diverse tasks, even when heavily overparameterized. Existing works offer only partial explanations (for example, the NTK-based task-model alignment explanation neglects feature learning). Here, we provide an end-to-end, analytically tractable case study that links a network's inductive prior, its training dynamics including feature learning, and its eventual generalisation. Specifically, we exploit the one-to-one correspondence between depth-2 discrete fully connected networks and disjunctive normal form (DNF) formulas by training on Boolean functions. Under a Monte Carlo learning algorithm, our model exhibits predictable training dynamics and the emergence of interpretable features. This framework allows us to trace, in detail, how inductive bias and feature formation drive generalisation.

Previous design methodologies for spring-driven jumping robots focused on jump height optimization for specific tasks. In doing so, numerous designs have been proposed including using nonlinear spring-linkages to increase the elastic energy storage and jump height. However, these systems can never achieve their theoretical maximum jump height due to taking off before the spring energy is fully released, resulting in an incomplete transfer of stored elastic energy to gravitational potential energy. This paper presents low-order models aimed at characterising the energy conversion during the acceleration phase of jumping. It also proposes practical solutions for increasing the energy efficiency of jumping robots. A dynamic analysis is conducted on a multibody system comprised of rotational links, which is experimentally validated using a physical demonstrator. The analysis reveals that inefficient energy conversion is attributed to inertial effects caused by rotational and unsprung masses. Since these masses cannot be entirely eliminated from a physical linkage, a practical approach to improving energy efficiency involves structural redesign to reduce structural mass and moments of inertia while maintaining compliance with structural strength and stiffness requirements.

When deploying machine learning (ML) models on embedded and IoT devices, performance encompasses more than an accuracy metric: inference latency, energy consumption, and model fairness are necessary to ensure reliable performance under heterogeneous and resource-constrained operating conditions. To this end, prior research has studied model-centric approaches, such as tuning the hyperparameters of the model during training and later applying model compression techniques to tailor the model to the resource needs of an embedded device. In this paper, we take a data-centric view of embedded ML and study the role that pre-processing parameters in the data pipeline can play in balancing the various performance metrics of an embedded ML system. Through an in-depth case study with audio-based keyword spotting (KWS) models, we show that pre-processing parameter tuning is a remarkable tool that model developers can adopt to trade-off between a model's accuracy, fairness, and system efficiency, as well as to make an embedded ML model resilient to unseen deployment conditions.