Vertical displacement events (VDEs) can occur in elongated tokamaks causing large currents to flow in the vessel and other adjacent metallic structures. To better understand the potential magnitude of the associated forces and the role of the so called ``halo currents'' on them, we have used the M3D-C1 code to simulate potential VDEs in ITER. We used actual values for the vessel resistivity and pre-quench temperatures and, unlike most of the previous studies, the halo region is naturally formed by triggering the thermal quench with an increase in the plasma thermal conductivity. We used the 2D non-linear version of the code and vary the post-thermal quench thermal conductivity profile as well as the boundary temperature in order to generate a wide range of possible cases that could occur in the experiment. We also show that, for a similar condition, increasing the halo current does not increase the total force on the wall since it is offset by a decrease in the toroidal contribution.
Extrapolation -- the ability to make inferences that go beyond the scope of one's experiences -- is a hallmark of human intelligence. By contrast, the generalization exhibited by contemporary neural network algorithms is largely limited to interpolation between data points in their training corpora. In this paper, we consider the challenge of learning representations that support extrapolation. We introduce a novel visual analogy benchmark that allows the graded evaluation of extrapolation as a function of distance from the convex domain defined by the training data. We also introduce a simple technique, context normalization, that encourages representations that emphasize the relations between objects. We find that this technique enables a significant improvement in the ability to extrapolate, considerably outperforming a number of competitive techniques.