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.
Yoo, Jongsoo; Jara-almonte, J.; Yerger, Evan; Wang, Shan; Qian, Tony; Le, Ari; Ji, Hantao; Yamada, Masaaki; Fox, William; Kim, Eun-Hwa; Chen, Li-Jen; Gershman, Daniel
Whistler wave generation near the magnetospheric separatrix during reconnection at the dayside magnetopause is studied with data from the Magnetospheric Multiscale (MMS) mission. The dispersion relation of the whistler mode is measured for the first time near the reconnection region in space, which shows that whistler waves propagate nearly parallel to the magnetic field line. A linear analysis indicates that the whistler waves are generated by temperature anisotropy in the electron tail population. This is caused by loss of electrons with a high velocity parallel to the magnetic field to the exhaust region. There is a positive correlation between activities of whistler waves and the lower-hybrid drift instability (LHDI) both in laboratory and space, indicating the enhanced transport by LHDI may be responsible for the loss of electrons with a high parallel velocity.