Artificial Neural Network
- Nitish Kumar Sharma
- 13000116090
Index
1. Early Neural Networks
2. Perceptron
3. Sigmoid Neurons
4. Architecture of ANN
5. Feed Forward Neural Network
6. Recurrent Neural Network
7. Current Study
8. Applications
Early Neural Networks
1. Early ideas were investigated as early as 1950s.
2. In 1980’s a lot of people discovered Backpropagation independently
using gradient-descent optimization and started applying to neural
networks.
3. First successful implementation of Neural Network in 1989 in Bell
Labs.
4. Yan LeCun combined earlier ideas of Convolutional network with
Backpropagation and was applied to the problem of classifying
handwritten digits.
5. Resulting network was used by United States Postal Service in
1990s to automate reading of ZIP codes on mail envelopes.
Perceptron
1. A perceptron takes several binary inputs x1,x2,.... and produces a
single binary output.
2. Rosenblatt introduced weights w1,w2,.... real numbers expressing
the importance of the respective inputs to the output.
3. In algebraic terms the perceptron works by the equation:
Perceptron
1. In simple form w,x to represent weight and inputs respectively and
b is the perceptron bias. The equation is:
Sigmoid Neurons
1. In this case the inputs can take any value between 0 & 1. So for
example 0.76 is a valid input for sigmoid neuron.
2. The output here is σ(w.x+b),where σ is called logistic function and is
defined by:
Here z will be replaced by (w.x+b).
Architecture of ANN
1. Leftmost layer is called input layer. Rightmost layer is called output
layer. Middle layers are called hidden layers.
Feed Forward Neural Network
1. Feed Forward Topology with only one condition: information must
flow from input to output in only one direction with no back-loops.
2. A simple feed forward neural network
Recurrent Neural Network
1. Information is also transmitted backward.
2. Can use internal memory to process any sequence of inputs.
Current Study
1. Network Intrusion Detection using Deep Neural Network: The deep
neural networks have a higher detection rate (94.57%) and
accuracy (82.78%) and a lower false positive rate (4.72%) as
compared to the corresponding average values of 89.03%,77.53%
and 10.95% from classical machine learning.
2. Accelerated learning in a Physical-Model Spiking Neural Network:
Applications
1. Fuzzy logic and neural networks have been integrated for uses as
diverse as automotive engineering, applicant screening for jobs, the
control of a crane, and monitoring glaucoma.
2. Prediction of Meteorological Parameters during Pre Monsoon
Thunderstorms.
3. In Health Care Diagnosis and Management System.
4. Encoding neural network devices onto chip or specialised analog
devices for speed.
5. ANNs are often used in silico drug design for correlation,
classification, and predictions of the activity of bioactive compounds
and drug.
References
➔ Introduction to the Artificial Neural Networks Andrej Krenker1, Janez
Bešter2 and Andrej Kos2
➔ [Link] [Free Online book]
➔ Simon Friedmann et al. “Demonstrating Hybrid Learning in a
Flexible Neuromorphic Hardware System”. In: IEEE Transactions on
Biomedical Circuits and Systems 11.1 (Feb. 2017), pp. 128–142. doi:
10.1109/TBCAS. 2016.2579164.
➔ Timo Wunderlich et al. “Demonstrating Advantages of Neuromorphic
Computation: A Pilot Study”. In: Frontiers in Neuroscience (2019).
doi: 10. 3389/fnins.2019.00260.
References
➔ Sacha J. van Albada et al. “Performance Comparison of the Digital
Neuromorphic Hardware SpiNNaker and the Neural Network
Simulation Software NEST for a Full-Scale Cortical Microcircuit
Model”. In: Frontiers in Neuroscience 12 (May 2018), p. 291. issn:
1662-453X. doi: 10.3389/fnins. 2018.00291.
➔ LuNet: A Deep Neural Network for Network Intrusion Detection
Peilun Wu∗ and Hui Guo†
➔ [Link]
➔ Genetic Algorithm Optimization of Bayesian-Regularized Artificial
Neural Networks in Drug Design MichaelFernandez1JulioCaballero2
➔ Deep Learning With Python:Francois Chollet.