This paper describes the design and simulation of a proof-of-concept quasi-integrable octupole lattice at the University of Maryland Electron Ring (UMER). This experiment tests the feasibility of nonlinear integrable optics, a novel technique that is expected to mitigate resonant beam loss and enable low-loss high-intensity beam transport in rings. Integrable lattices with large amplitudedependent tune spreads, created by nonlinear focusing elements, are proposed to damp beam response to resonant driving perturbations while maintaining large dynamic aperture [Danilov and Nagaitsev, PRSTAB, 2010]. At UMER, a lattice with a single octupole insert is designed to test the predictions of this theory. The planned experiment employs a low-current high-emittance beam with low space charge tune shift (∼ 0.005) to probe the dynamics of a lattice with large externally-induced tune spread. Design studies show that a lattice composed of a 25-cm octupole insert and existing UMER optics can induce a tune spread of ∼ 0.13. Stable transport is observed in PIC simulation for many turns at space charge tune spread 0.008. A maximum spread of ∆ν = 0.11 (RMS 0.015) is observed for modest octupole strength (peak 50 T /m 3 ). A simplified model of the system explores beam sensitivity to steering and focusing errors. Results suggest that control of orbit distortion to < 0.2 mm within the insert region is essential. However, we see only weak dependence on deviations of lattice phase advance (≤ 0.1 rad.) from the invariant-conserving condition.

No abstract

We describe the continuous-time dynamics of networks implemented on Field Programable Gate Arrays (FPGAs). The networks can perform Boolean operations when the FPGA is in the clocked (digital) mode; however, we run the programed FPGA in the unclocked (analog) mode. Our motivation is to use these FPGA networks as ultrafast machine-learning processors, using the technique of reservoir computing. We study both the undriven dynamics and the input response of these networks as we vary network design parameters, and we relate the dynamics to accuracy on two machine-learning tasks.

No abstract

Despite state-of-the-art deep learning-based computer vision models achieving high accuracy on object recognition tasks, x-ray screening of baggage at checkpoints is largely performed by hand. Part of the challenge in automation of this task is the relatively small amount of available labeled training data. Furthermore, realistic threat objects may have forms or orientations that do not appear in any training data, and radiographs suffer from high amounts of occlusion. Using deep generative models, we explore data augmentation techniques to expand the intra-class variation of threat objects synthetically injected into baggage radiographs using openly available baggage x-ray datasets. We also benchmark the performance of object detection algorithms on raw and augmented data.