ITR: Quantum Information Processing with Ultracold Rydberg
Atoms
Robin Côté, Edward Eyler, and Phillip Goul
The goal of this research at University of Connecticut was to investigate
the feasibility of quantum information processing using ultracold Rydberg
atoms, at both theoretical and experimental levels. The short-term goal
was to demonstrate that entangled states could be generated with Rydberg
atoms, and the longer term goal was to realize a quantum logic gate, namely
the phase gate.
By selecting the Rydberg states judiciously, researchers found that it
may be possible to obtain a conditional excitation where only one atom
can be excited at a time. We showed that this effect, named dipole blockade,
could be generalized to an ensemble of atoms, i.e., only one atom can
be excited into a Rydberg state in the mesoscopic ensemble. Such an effect
would simplify the manipulation of information and could be used to entangle
various ensembles.
Adaptive Neurally-Inspired Computing: Models,
Algorithms, and Silicon-Based Architectures
Rajesh P. N. Rao and Christopher J. Diorio
The goal of this research at the University of Washington was to discover
primitives for adaptive neural computing by simulating biological neurons
and synapses, develop algorithms for computation and learning in networks
of spiking neurons using these primitives, and implement these algorithms
in silicon using synapse transistors and field-programmable learning arrays
(FPLAs).
A new learning algorithm was developed based on the recently discovered
neurobiological phenomenon known as redistribution of synaptic efficacy.
The learning algorithm adapted the input connections to a neuron in such
a way that the neuron became selective for specific temporal patterns.
This allowed a network of neurons to learn temporal sequences (such as
audio or video) without any external supervision.
This research used a Matlab-based simulator for investigating computing
and learning in networks of spiking (integrate-and-fire) neurons.
Primate-Inspired Specialized Learning in
an Agent Architecture: Safe, Robust, Adaptive Action Selection
Marc D. Hauser (PI) and Joanna Bryson
This pilot study (SGER) by researchers at Harvard University explores whether
modeling the ways non-human primates learn can improve safe learning in artificial
agents and seeks
to better understand the limits of learning that primates (sometimes including
humans) have. These limits may actually be useful for agents (both natural and
artificial) because they help make learning quick and accurate. They may also
safeguard learning from interfering destructively with existing knowledge.
Eight months into the grant, the PIs had five related software-agent
workshop publications accepted and an animal-cognition journal article
in preparation. Current simulations are developing entirely new models
of task-learning in primates. The research has been accelerated by Jonathon
Leong, a neuroscience undergraduate, who has been running experiments,
and, as a side effect, contributing to the debugging of the software environment
being used for building the agents. Producing educational tools is a long-term
goal of the project.
|
