Abstract: Current AI visual algorithms are very limited compared to the robustness and flexibility of the human visual system. These limitations, however, are often obscured by the standard performance measures (SPMs) used to evaluate vision algorithms which favor data-driven methods. SPMs, however, are problematic due to the combinatorial complexity of natural images and lead to unrealistic expectations about the effectiveness of current algorithms. We argue that tougher performance measures, such as out-of-distribution testing and adversarial examiners, are required to realistically evaluate vision algorithms and hence to encourage AI vision systems which can achieve human level performance. We illustrate this by studying object classification where the algorithms are trained on standard datasets which have limited occlusion but are tested on datasets where the objects are severally occluded (out-of-distribution testing) and/or where adversarial patches are placed in the images (adversarial examiners). We show that standard Deep Nets perform badly under these types of tests but Generative Compositional Nets, which perform approximate analysis by synthesis, are much more robust.

Zoom link: https://mit.zoom.us/j/95505708173?pwd=cjBLVlZWYXNXcDBIanRKMWZNNXZuZz09
Passcode: 522130

Abstract: About fifty years ago, holography was proposed as a model of associative memory. Associative memories with similar properties were soon after implemented as simple networks of threshold neurons by Willshaw and Longuet-Higgins. It turns out that the recurrent Willshaw networks were very similar to today's deep nets. Thinking about deep learning in terms of associative networks memories a more realistic and sober perspective on the promises of deep learning and on its role in eventually understanding human intelligence.

This panel discussion will be hosted remotely via Zoom.

Zoom link: https://mit.zoom.us/j/99556310473

Host: Prof. Tomaso Poggio (MIT)

Abstract:  Despite great advances in ML, and in our understanding of the brain at the level of neurons, synapses, and neural circuits, we still have no satisfactory explanation for the brain's performance in perception, cognition, language, memory, behavior; as Nobel laureate Richard Axel put it, ``we have no logic for translating neural activity into thought and action''. The Assembly Calculus (AC) is a framework to fill this gap, a computational model whose basic data type is the assembly, a large subset of neurons whose simultaneous excitation is tantamount to the subject's thinking of an object, idea, episode, or word. The AC provides a repertoire of operations ("project", "reciprocal-project", "associate", "pattern-complete", etc.) whose implementation relies only on Hebbian plasticity and inhibition, and encompasses a complete computational system, thereby enabling complex function. Very recently, it has been shown, rigorously and in simulation, that the AC can learn to classify samples from well-separated classes. For basic concept classes in high dimension, an assembly can be formed and recalled for each class, and these assemblies are distinguishable as long as the input classes are sufficiently separated. Viewed as a learning algorithm, this mechanism is entirely online, generalizes from very few samples, and requires only mild supervision --- all attributes expected of a brain-like mechanism. The talk will highlight several fascinating questions that arise, from the convergence of assemblies to their unexpected generalization abilities.

This is joint work with Christos Papadimitriou, Max Dabagia, Mirabel Reid and Dan Mitropolsky. ​

Zoom link: https://mit.zoom.us/j/97301534627

Abstract: In the last few years, the ability for robots to understand and operate in the world around them has advanced considerably. Examples include the growing number of self-driving car systems, the considerable work in robot mapping, and the growing interest in home and service robots. However, one limitation is that robots most often reason and plan using very geometric models of the world, such as point features, dense occupancy grids and action cost maps. To be able to plan and reason over long length and timescales, as well as planning more complex missions, robots need to be able to reason about abstract concepts such as landmarks, segmented objects and tasks (among other representations). I will talk about recent work in joint reasoning about semantic representations and physical representations and what these joint representations mean for planning and decision making.

This seminar series talk will be hosted remotely via Zoom.

Zoom link: https://mit.zoom.us/j/96323330576