Phase Transitions in Transfer Learning for High-Dimensional Perceptrons

Transfer learning seeks to improve the generalization performance of a target task by exploiting the knowledge learned from a related source task. Central questions include deciding what information one should transfer and when transfer can be beneficial. The latter question is related to the so-called negative transfer phenomenon, where the transferred source information actually reduces the generalization performance of the target task. This happens when the two tasks are sufficiently dissimilar. In this paper, we present a theoretical analysis of transfer learning by studying a pair of related perceptron learning tasks. Despite the simplicity of our model, it reproduces several key phenomena observed in practice. Specifically, our asymptotic analysis reveals a phase transition from negative transfer to positive transfer as the similarity of the two tasks moves past a well-defined threshold.

Synaptic plasticity in distal dendrites can optimize probabilistic top-down predictions in pyramidal cells

Predictive coding has been identified as a major driver of computation and learning in corticalmicrocircuits. But it has remained unknown which synaptic plasticity processes install and maintain predictive coding capability. Predictions are inherently uncertain, and learning rules that aim at discriminating linearly separable classes of inputs - such as the perceptron learning rule - are not suitable for learning to predict. We show that experimental data on synaptic plasticity in distal dendrites of pyramidal cells support a functionally more powerful learning rule that produces a spike-based approximation to logistic regression, i.e., to probabilistic prediction. We also show that experimentally found interactions between different dendritic branches support learning of predictions for more complex probability distributions. The resulting learning theory for top-down inputs to pyramidal cells provides a normative framework for evaluating experimental data, and suggests further experiments for tracking the emergence of predictive coding through synaptic plasticity in distal dendrites.

A technology for prediction of preterm discharge of amniotic fluid

Aim of study. To create a technology for prediction of preterm discharge of amniotic fl uid based on universally accessible methods of laboratory and instrumental evaluation. Material and methods. A retrospective analysis of 200 birth cases dated 2018-2021 at the premises of obstetric facilities in Chita and Ufa cities featuring patients admitted to the inpatient unit shortly before term labour (1-2 days). In the course of the study, 2 groups were distinguished: Group 1 included 128 female patients with term discharge of amniotic fluid and Group 2 was constituted by 72 female patients with preterm discharge of amniotic fluid. The groups were comparable in age, anthropomorphic parameters and extragenital pathology. On admission, all women underwent general medical examination and ultrasonography. Statistical processing of the results was performed via the IBM SPSS Statistics Version 25.0 soft ware. Results. The technology for prediction of preterm discharge of amniotic fluid was based on multilayer perceptron learning. The structure of the learning neural network included 5 input neurons: body mass index, fundal height, the total bilirubin level, activated partial thromboplastin time and the amniotic fluid index. Th e percentage of incorrect predictions of the neural network totalled 28.5 %. Conclusion. A complex approach based on integration of universally accessible methods for laboratory and instrumental tests shortly before the labour based on a neural network makes it possible to predict possible preterm discharge of amniotic fl uid with an accuracy of up to 75 %. Application of this technology in clinical practice will make it possible not only to perform timely preparation of the parturient canal but also to reduce the frequency of adverse obstetric and perinatal outcomes

MARINE FIRE-DANGEROUS SITUATIONS FACTORS’ VALUES FORECASTING BASED ON ARTIFICIAL NEURAL NETWORK

The article describes the procedure of marine fire-dangerous situations factors’ values forecasting based on artificial neural network. These factors are temperature, optical air density, aerosol concentration. Given procedure is flexible and can be expanded for other factors of fire-safety state of monitored object. Artificial neural network with architecture of three-layer perceptron is used for forecasting. The article gives a common scheme for realization of fire-dangerous situations factors’ values forecasting, substantiates the choice of used artificial neural network’s architecture, gives perceptron learning algorithm. As a result of given procedure execution factors’ values forecasting is implemented for prevention of fire-dangerous situation and the adoption of early actions. In case of integration of the developed procedure inside ship information management systems’ algorithmic support is capable of dramatically raise effectiveness of decisions made while providing fire safety on ships.

Alphanumeric Pattern Recognition by Memristive Crossbar Circuit using Perceptron Learning Rule

This paper presents an alphanumeric pattern recognition approach based on memristive crossbar circuit using perceptron learning rule. The proposed approach incorporates a memristive crossbar-based learning and training circuit (TC) module (i.e., synaptic network) and an operational amplifier (op-amp)-based neuron. Alphanumeric patterns, such as alphabets (A–Z) and numerics (0–9), are applied on the TC module and it adjusts the synaptic weights using the perceptron learning rule. The TC module includes 16 inputs, which are interconnected to nine output neurons through memristors. The input and output patterns are represented through [Formula: see text] and [Formula: see text] matrix pixels, respectively. This proposed circuit has implemented all alphanumeric patterns, such as alphabets (A–Z) and numerics (0–9), successfully. However, only the pattern “A” is illustrated in detail for better understanding. SPICE simulation results supported by analytical calculations of pattern “A” are reported. The average power consumption for the proposed approach using memristor is 77.77% lower than the conventional MOSFET-based approach, apart from significant saving of silicon overhead in contrast to its counterpart approach.