No abstract

No abstract

Performance characteristics of a new design of positron tomograph with automatically retractable septa for brain imaging have been studied. The device, consisting of block BGO detectors (8 x 8 elements per block), has a ring diameter of 76 cm and an axial FOV of 106.5 mm. The in-plane resolution is on average 5.8 mm and 5.0 mm (FWHM) for stationary and wobble sampling, respectively, over the central 18 cm of the transaxial FOV. Its unique feature is the capability of data acquisition both in the 'conventional' 2D mode (with septa) or 3D mode (septa retracted) where coincidences between any of the 16 detector rings are acquired. When scattered events are subtracted, the efficiency for a 20 cm diameter uniform cylinder increases overall by a factor of 4.8 between 2D (septa extended) and 3D modes. For a 20 cm phantom the trues/singles ratio is higher for 3D than for 2D but for a given unscattered trues rate, the randoms rate in 3D is higher. At 380 keV the scatter fraction within a 20 cm cylinder is 10% (septa extended) and 36% (retracted). In spite of the increase in scatter when septa are retracted, the increased efficiency in the 3D mode of acquisition yields distinct advantages, particularly in the many studies where tracer concentration is low and consequently where dead time and random rates are less important.

Recurrent neural networks (RNNs) can learn to perform finite state computations. It is shown that an RNN performing a finite state computation must organize its state space to mimic the states in the minimal deterministic finite state machine that can perform that computation, and a precise description of the attractor structure of such systems is given. This knowledge effectively predicts activation space dynamics, which allows one to understand RNN computation dynamics in spite of complexity in activation dynamics. This theory provides a theoretical framework for understanding finite state machine (FSM) extraction techniques and can be used to improve training methods for RNNs performing FSM computations. This provides an example of a successful approach to understanding a general class of complex systems that has not been explicitly designed, e.g., systems that have evolved or learned their internal structure.

No abstract