Parallel distributed processing and neural networks: origins, methodology and cognitive functions

Bridging the Finger-Action Gap between Hand Patients and Healthy People in Daily Life with a Biomimetic System

The hand is involved very deeply in our lives in daily activities. When a person loses some hand function, their life can be greatly affected. The use of robotic rehabilitation to assist patients in performing daily actions might help alleviate this problem. However, how to meet individual needs is a major problem in the application of robotic rehabilitation. A biomimetic system (artificial neuromolecular system, ANM) implemented on a digital machine is proposed to deal with the above problems. Two important biological features (structure-function relationship and evolutionary friendliness) are incorporated into this system. With these two important features, the ANM system can be shaped to meet the specific needs of each individual. In this study, the ANM system is used to help patients with different needs perform 8 actions similar to those that people use in everyday life. The data source of this study is our previous research results (data of 30 healthy people and 4 hand patients performing 8 activities of daily life). The results show that while each patient's hand problem is different, the ANM can successfully translate each patient's hand posture into normal human motion. In addition, the system can respond to this difference smoothly rather than dramatically when the patient's hand motions vary both temporally (finger motion sequence) and spatially (finger curvature).

Modeling the trend of coronavirus disease 2019 and restoration of operational capability of metropolitan medical service in China: a machine learning and mathematical model-based analysis

Background

To contain the outbreak of coronavirus disease 2019 (COVID-19) in China, many unprecedented intervention measures are adopted by the government. However, these measures may interfere in the normal medical service. We sought to model the trend of COVID-19 and estimate the restoration of operational capability of metropolitan medical service in China.

Methods

Real-time data of COVID-19 and population mobility data were extracted from open sources. SEIR (Susceptible, Exposed, Infectious, Recovered) and neural network models (NNs) were built to model disease trends in Wuhan, Beijing, Shanghai and Guangzhou. Combined with public transportation data, Autoregressive Integrated Moving Average (ARIMA) model was used to estimate the accumulated demands for nonlocal hospitalization during the epidemic period in Beijing, Shanghai and Guangzhou.

Results

The number of infected people and deaths would increase by 45% and 567% respectively, given that the government only has implemented traffic control in Wuhan without additional medical professionals. The epidemic of Wuhan (measured by cumulative confirmed cases) was predicted to reach turning point at the end of March and end in later April, 2020. The outbreak in Beijing, Shanghai and Guangzhou was predicted to end at the end of March and the medical service could be fully back to normal in middle of April. During the epidemic, the number of nonlocal inpatient hospitalizations decreased by 69.86%, 57.41% and 66.85% in Beijing, Shanghai and Guangzhou respectively. After the end of epidemic, medical centers located in these metropolises may face 58,799 (95% CI 48926–67,232) additional hospitalization needs in the first month.

Conclusion

The COVID-19 epidemic in China has been effectively contained and medical service across the country is expected to return to normal in April. However, the huge unmet medical needs for other diseases could result in massive migration of patients and their families, bringing tremendous challenges for medical service in major metropolis and disease control for the potential asymptomatic virus carrier.

Sequential Nonlinear Learning for Distributed Multiagent Systems via Extreme Learning Machines

We study online nonlinear learning over distributed multiagent systems, where each agent employs a single hidden layer feedforward neural network (SLFN) structure to sequentially minimize arbitrary loss functions. In particular, each agent trains its own SLFN using only the data that is revealed to itself. On the other hand, the aim of the multiagent system is to train the SLFN at each agent as well as the optimal centralized batch SLFN that has access to all the data, by exchanging information between neighboring agents. We address this problem by introducing a distributed subgradient-based extreme learning machine algorithm. The proposed algorithm provides guaranteed upper bounds on the performance of the SLFN at each agent and shows that each of these individual SLFNs asymptotically achieves the performance of the optimal centralized batch SLFN. Our performance guarantees explicitly distinguish the effects of data- and network-dependent parameters on the convergence rate of the proposed algorithm. The experimental results illustrate that the proposed algorithm achieves the oracle performance significantly faster than the state-of-the-art methods in the machine learning and signal processing literature. Hence, the proposed method is highly appealing for the applications involving big data.

Interconnected large-scale systems for three fundamental cognitive tasks revealed by functional MRI

The specific brain areas required to execute each of three fundamental cognitive tasks - object naming, same-different discrimination, and integer computation - are determined by whole-brain functional magnetic resonance imaging (fMRI) using a novel technique optimized for the isolation of neurocognitive systems. This technique (1) conjoins the activity associated with identical or nearly identical tasks performed in multiple sensory modalities (conjunction) and (2) isolates the activity conserved across multiple subjects (conservation). Cortical regions isolated by this technique are, thus, presumed associated with cognitive functions that are both distinguished from primary sensory processes and from individual differences. The object-naming system consisted of four brain areas: left inferior frontal gyrus, Brodmann's areas (BAs) 45 and 44; left superior temporal gyrus, BA 22; and left medial frontal gyrus, BA 6. The same-different discrimination system consisted of three brain areas: right inferior parietal lobule, BA 40; right precentral gyrus, BA 6; and left medial frontal gyrus, BA 6. The integer computation system consisted of five brain areas: right middle frontal gyrus, BA 6; right precentral gyrus, BA 6; left inferior parietal lobule, BA 40; left inferior frontal gyrus, BA 44; and left medial frontal gyrus, BA 6. All three neurocognitive systems shared one common cortical region, the left medial frontal gyrus, the object-naming and integer computation systems shared the left inferior frontal gyrus, and the integer computation and same-different discrimination systems shared the right precentral gyrus. These results are consistent with connectionist models of cognitive processes where specific sets of remote brain areas are assumed to be transiently bound together as functional units to enable these functions, and further suggest a superorganization of neurocognitive systems where single brain areas serve as elements of multiple functional systems.

Parallel distributed processing and executive functioning: Tower of Hanoi neural networkmodel in healthy controls and left frontal lobe patients

Previous clinical studies have demonstrated that executive functioning has a significant association with the frontal lobes. In this study we examined executive functioning with parallel distributed processing (PDP) methodology by attempting to replicate Tower of Hanoi (TOH) neuropsychological test performance differences in normal healthy controls vs. left frontal lobe patients. Initial theoretical experiments were carried out to solve computer subprocessing constraints of the TOH partially by determining the programming prerequisites for a specific number of rings, efficient sequencing, and a general learning rule. These preliminary studies were then modified using a back-propagation neurocomputational approach. Computer program output showed a resemblance to normal controls' TOH moves. Degrading of the left frontal lobe patients' network resulted in characteristic impairments of executive functioning, i.e., perseveration and rule-braking.

Increased regional cerebral glucose metabolism and semantic memory performance in Alzheimer's disease: a pilot double blind transdermal nicotine positron emission tomography study

Nicotinic receptor dysfunction and impaired semantic memory occur early in Alzheimer's disease patients (AD). Previous research implied that nicotine's ability to enhance alertness, arousal, and cognition in a number of nonclinical populations was a function of its ability to stimulate CNS nicotinic cholinergic receptors. In this study it was hypothesized that transdermal administration of nicotine would increase both regional cerebral glucose metabolism (rCMRglc) and semantic memory (as assessed by verbal fluency). Two mild AD and two elderly controls underwent positron emission tomography scanning during a double blind nicotinic agonist verbal fluency challenge procedure. rCMRglc increases occurred in both AD patients, but not controls. In the two AD patients, verbal fluency scores increased by an average of 17%. One elderly control's verbal fluency increased, and the other decreased. These findings suggest that nicotine's effect on metabolism and verbal fluency is due to its ability to stimulate the cholinergic system.

Methodological and theoretical issues in neural network models of frontal cognitive functions

Neural network models have made significant strides in recent years toward modeling of neuropsychological data. In particular, three different research groups, including that of the present authors, have simulated in model networks some of the behavioral effects of frontal lobe damage. In this article we review these models and discuss their significance in terms of hierarchical organization in the nervous system. These models, to varying degrees, incorporate several widely used neural network principles that have also been used to model a wide range of data in other areas such as categorization, conditioning, and motor control. These principles include associative learning, competition, opponent processing, neuromodulation, and interlevel resonant feedback. Specifically, we show how combinations of these principles serve to model the attentional requirements of cognitive tasks and motor plans.

Parallel distributed processing and neuropsychology: a neural network model of Wisconsin Card Sorting and verbal fluency

Neural networks can be used as a tool in the explanation of neuropsychological data. Using the Hebbian Learning Rule and other such principles as competition and modifiable interlevel feedback, researchers have successfully modeled a widely used neuropsychological test, the Wisconsin Card Sorting Test. One of these models is reviewed here and extended to a qualitative analysis of how verbal fluency might be modeled, which demonstrates the importance of accounting for the attentional components of both tests. Difficulties remain in programming sequential cognitive processes within a parallel distributed processing (PDP) framework and integrating exceedingly complex neuropsychological tests such as Proverbs. PDP neural network methodology offers neuropsychologists co-validation procedures within narrowly defined areas of reliability and validity.