ABSTRACT
Despite the advancement of machine learning techniques in recent years, state-of-the-art systems lack robustness to "real world" events, where the input distributions and tasks encountered by the deployed systems will not be limited to the original training context, and systems will instead need to adapt to novel distributions and tasks while deployed. This critical gap may be addressed through the development of "Lifelong Learning" systems that are capable of (1) Continuous Learning, (2) Transfer and Adaptation, and (3) Scalability. Unfortunately, efforts to improve these capabilities are typically treated as distinct areas of research that are assessed independently, without regard to the impact of each separate capability on other aspects of the system. We instead propose a holistic approach, using a suite of metrics and an evaluation framework to assess Lifelong Learning in a principled way that is agnostic to specific domains or system techniques. Through five case studies, we show that this suite of metrics can inform the development of varied and complex Lifelong Learning systems. We highlight how the proposed suite of metrics quantifies performance trade-offs present during Lifelong Learning system development - both the widely discussed Stability-Plasticity dilemma and the newly proposed relationship between Sample Efficient and Robust Learning. Further, we make recommendations for the formulation and use of metrics to guide the continuing development of Lifelong Learning systems and assess their progress in the future.
Subject(s)
Education, Continuing , Machine LearningABSTRACT
The koniocellular (K) layers of the primate dorsal lateral geniculate nucleus house a variety of visual receptive field types, not all of which have been fully characterized. Here we made single-cell recordings targeted to the K layers of diurnal New World monkeys (marmosets). A subset of recorded cells was excited by both increments and decrements of light intensity (on/off-cells). Histological reconstruction of the location of these cells confirmed that they are segregated to K layers; we therefore refer to these cells as K-on/off cells. The K-on/off cells show high contrast sensitivity, strong bandpass spatial frequency tuning, and their response magnitude is strongly reduced by stimuli larger than the excitatory receptive field (silent suppressive surrounds). Stationary counterphase gratings evoke unmodulated spike rate increases or frequency-doubled responses in K-on/off cells; such responses are largely independent of grating spatial phase. The K-on/off cells are not orientation or direction selective. Some (but not all) properties of K-on/off cells are consistent with those of local-edge-detector/impressed-by-contrast cells reported in studies of cat retina and geniculate, and broad-thorny ganglion cells recorded in macaque monkey retina. The receptive field properties of K-on/off cells and their preferential location in the ventral K layers (K1 and K2) make them good candidates for the direct projection from geniculate to extrastriate cortical area MT/V5. If so, they could contribute to visual information processing in the dorsal ("where" or "action") visual stream.SIGNIFICANCE STATEMENT We characterize cells in an evolutionary ancient part of the visual pathway in primates. The cells are located in the lateral geniculate nucleus (the main visual afferent relay nucleus), in regions called koniocellular layers that are known to project to extrastriate visual areas as well as primary visual cortex. The cells show high contrast sensitivity and rapid, transient responses to light onset and offset. Their properties suggest they could contribute to visual processing in the dorsal ("where" or "action") visual stream.
