No abstract

Deep neural networks are typically too computationally expensive to run in real-time on consumer-grade hardware and low-powered devices. In this paper, we investigate reducing the computational and memory requirements of neural networks through network pruning and quantisation. We examine their efficacy on large networks like AlexNet compared to recent compact architectures: ShuffleNet and MobileNet. Our results show that pruning and quantisation compresses these networks to less than half their original size and improves their efficiency, particularly on MobileNet with a 7× speedup. We also demonstrate that pruning, in addition to reducing the number of parameters in a network, can aid in the correction of overfitting.

No abstract

Graph neural networks (GNNs) build on the success of deep learning models by extending them for use in graph spaces. Transfer learning has proven extremely successful for traditional deep learning problems, resulting in faster training and improved performance. Despite the increasing interest in GNNs and their use cases, there is little research on their transferability. This research demonstrates that transfer learning is effective with GNNs, and describes how source tasks and the choice of GNN impact the ability to learn generalisable knowledge. We perform experiments using real-world and synthetic data within the contexts of node classification and graph classification. To this end, we also provide a general methodology for transfer learning experimentation and present a novel algorithm for generating synthetic graph classification tasks. We compare the performance of GCN, GraphSAGE and GIN across both synthetic and real-world datasets. Our results demonstrate empirically that GNNs with inductive operations yield statistically significantly improved transfer. Further, we show that similarity in community structure between source and target tasks support statistically significant improvements in transfer over and above the use of only the node attributes.

Variations in build process parameters, post-processing parameters, and feedstock have a significant impact on the structural integrity and performance of components made with additive manufacturing (AM). Effective nondestructive testing (NDT) is critical for ensuring the structural integrity of components. Complex geometries, nonequilibrium microstructures, new process variables, and lack of clear accept or reject criteria for AM components present new challenges to NDT. Quantitative, volumetric NDT methods that can detect material defects of interest in complex geometries are required. Process compensated resonance testing (PCRT) is an NDT method that uses a swept sine input to excite the component's resonance modes of vibration. The resonance frequencies are recorded, analyzed statistically, and compared to acceptability limits established using a database of training components. The swept sine input excites whole-body vibrational modes in nearly any geometry, and the component's resonance frequencies correlate directly to its structural integrity. In this study, PCRT evaluations were performed on titanium alloy (Ti-6Al-4V) populations made with electron beam PBF and aluminum alloy (AlSi10Mg) populations made with laser PBF. The evaluations were conducted in support of ASTM round-robin testing. In the Ti-6Al-4V population, PCRT showed clear resonance frequency differences between nominal specimens and off-nominal specimens with defective material states. PCRT also quantified the effects of hot isostatic pressing (HIP). PCRT pass/fail NDT of the Ti-6Al-4V population in the pre-HIP and post-HIP states demonstrated 100% accuracy. Computed tomography scans of the post-HIP specimens showed no clear indications of porosity. Follow-up tensile testing of a subset of nominal and off-nominal specimens in the post-HIP state showed that the off-nominal specimens had lower yield stresses and ultimate tensile stresses than nominal specimens. In the AlSi10Mg population, PCRT detected differences between recycled and virgin feedstock powder. PCRT pass/fail NDT of AlSi10Mg specimens exposed to nominal and off-nominal heat treatment demonstrated 100% accuracy.