Automated discovery of algorithms from data.

To automate the discovery of new scientific and engineering principles, artificial intelligence must distill explicit rules from experimental data. This has proven difficult because existing methods typically search through the enormous space of possible functions. Here we introduce deep distilling, a machine learning method that does not perform searches but instead learns from data using symbolic essence neural networks and then losslessly condenses the network parameters into a concise algorithm written in computer code. This distilled code, which can contain loops and nested logic, is equivalent to the neural network but is human-comprehensible and orders-of-magnitude more compact. On arithmetic, vision and optimization tasks, the distilled code is capable of out-of-distribution systematic generalization to solve cases orders-of-magnitude larger and more complex than the training data. The distilled algorithms can sometimes outperform human-designed algorithms, demonstrating that deep distilling is able to discover generalizable principles complementary to human expertise.

Explainable neural networks that simulate reasoning.

The success of deep neural networks suggests that cognition may emerge from indecipherable patterns of distributed neural activity. Yet these networks are pattern-matching black boxes that cannot simulate higher cognitive functions and lack numerous neurobiological features. Accordingly, they are currently insufficient computational models for understanding neural information processing. Here, we show how neural circuits can directly encode cognitive processes via simple neurobiological principles. To illustrate, we implemented this model in a non-gradient-based machine learning algorithm to train deep neural networks called essence neural networks (ENNs). Neural information processing in ENNs is intrinsically explainable, even on benchmark computer vision tasks. ENNs can also simulate higher cognitive functions such as deliberation, symbolic reasoning and out-of-distribution generalization. ENNs display network properties associated with the brain, such as modularity, distributed and localist firing, and adversarial robustness. ENNs establish a broad computational framework to decipher the neural basis of cognition and pursue artificial general intelligence.

Dynamic pupillometry as an autonomic testing tool.

OBJECTIVE: To determine normal values for pupillometry indices in healthy control subjects and to examine these indices in patients with autonomic dysfunction and healthy controls. METHODS: Infrared video pupillometry was used to investigate the pupil response to a brief light flash in 79 healthy controls, 28 patients with normal autonomic function (composite autonomic severity score, CASS < 2), and 26 patients with moderate to severe autonomic failure (CASS > 4) seen in our autonomic laboratory from January 2008 to June 2011. In six subjects, we examined the effects of varying light stimulus intensity and light stimulus duration. Descriptive analysis, correlation, and ANCOVA adjusted for age were performed. RESULTS: We determined eight indices corresponding to parasympathetic and sympathetic pupil function. Baseline pupil diameter (BPD), maximum constriction velocity (MCV), absolute constriction amplitude (ACA), and maximum dilation velocity (MDV) negatively correlated with age (p < 0.01) among controls. MCV and ACA increased with increasing intensity of light stimulus from 3.5 to 112 µW. Indices of parasympathetic pupil innervation (MCV and ACA) were lower in the high CASS group compared to others (p < 0.0001). Indices of sympathetic pupil function, time to reach 75 % of initial resting diameter during pupillary dilation (T¾), and dilation velocity at T¾ (DV¾) did not differ significantly in the three study groups. However, T¾ corrected for the magnitude of pupillary constriction (T¾:ACA) was higher in the high CASS group suggesting sympathetic dysfunction in that group (p = 0.0003). CONCLUSIONS: Indices of pupillomotor function significantly differ between patients with moderate to severe autonomic failure and healthy controls.

Objective characterization of the relative afferent pupillary defect in MS.

OBJECTIVE: To develop an objective and precise neurophysiologic method from which to identify and characterize the presence and magnitude of relative afferent pupillary defects (RAPD) in patients with MS. METHODS: Binocular infrared pupillometry was performed in 40 control subjects and 32 MS patients with RAPDs, using two precisely defined sequences of alternating light flashes (right-left and left-right). We analyzed three distinct pupillary metrics in response to light stimulation. These included percent diameter change (DC), constriction curve area (CCA), which measures change in diameter over time, and the phase-plane curve area (PCA) which measures change in diameter with change in velocity. Direct and consensual response ratios (for each eye) were computed and analyzed for each metric in response to both the first flash (i.e. first phase) and second flash (i.e. second phase) of the 'swinging flashlight' test. RESULTS: Second flash pupillary response metric asymmetry ratios yielded the highest discriminatory power for RAPD detection. Receiver operating characteristic areas under the curve for each of the pupillary metric response asymmetry ratios were as follows: diameter change: 0.97; constriction curve area: 0.96; phase-plane curve area: 0.95 (p<0.0001 for all comparisons compared to normal subjects). The sum of these three squared ratios (SSR) yielded a combined metric with the greatest discriminatory power (receiver operator characteristic area under the curve=0.99). CONCLUSIONS: Second flash (i.e. the second phase of the swinging light test) pupillary metric response asymmetry ratios are highly sensitive and specific for the confirmation and characterization of an RAPD in patients with MS. This objective neurophysiologic method may be useful for studying the relationship between a stereotyped reflex, and nervous system architecture, with potential ramifications for detecting and monitoring neuroprotective and restorative effects of novel agents in MS treatment trials.