Digital Electronic Systems
BJT logic gates: RTL, DTL, TTL 7400, Schottky and ECL
The transistor as a switch and saturation, RTL NOR and fan-out proved, DTL, the TTL 7400 NAND state by state with its…
Category
Fourth Year
Fourth-year subjects: applied electronics, systems theory, electrical communications, digital electronics, electrical measurements and process control.
82 articles
Applied Electronics
Solved transistor exercises: biasing, small-signal amplifiers and logic gates
Twelve exercises with solutions: divider design, pnp, sensitivity to beta, JFET and MOSFET, common emitter with degeneration, follower, common source, Miller effect…
Applied Electronics
Small-signal amplifiers: π and T models, common emitter, base and collector, Miller effect
Transconductance and distortion proved, equivalent π and T models, gain and resistances of the BJT and FET configurations, a two-stage amplifier computed…
Machine Learning and Neural Networks
Project: recognising handwritten digits in NumPy and PyTorch, up to ONNX and 8-bit quantisation
A network written from scratch in NumPy and the same one in PyTorch on 8x8 digits (98 % accuracy), confusion matrix and…
Machine Learning and Neural Networks
Training in practice: data, He initialisation, normalisation, dropout and metrics
Training, validation and test, He initialisation proved and measured over 20 layers, batch and layer normalisation, dropout, amount of data, precision, recall…
Machine Learning and Neural Networks
Sequences: recurrent networks, vanishing gradient, LSTM, attention and transformers
Recurrent networks and backpropagation through time, the vanishing gradient proved and measured, the LSTM, attention with the proof of the division by…
Machine Learning and Neural Networks
Convolutional networks: convolution, translation equivariance, pooling and sizes
Two-dimensional convolution, translation equivariance proved, pooling, size and parameter-count formulas, LeNet, AlexNet and ResNet, dense network versus CNN on handwritten digits.
Machine Learning and Neural Networks
Optimisers: SGD, momentum, RMSProp and Adam with bias correction
Mini-batch stochastic gradient descent, momentum, RMSProp, Adam and the bias correction proved, AdamW, warm-up and step decay, comparison of the methods on…
Machine Learning and Neural Networks
Backpropagation: equations proved in matrix form and numerical gradient check
The chain rule on the computational graph, the backpropagation equations proved, cost independent of the number of parameters, automatic differentiation, from-scratch NumPy…
Machine Learning and Neural Networks
The perceptron and multilayer networks: Novikoff's theorem, XOR and universal approximation
The perceptron and Novikoff's convergence theorem proved, the XOR problem, multilayer networks and activation functions, the universal approximation theorem, a 100-seed experiment…