Bootcamp - Master Neural Network

Our company Sci-Code's mission is to provide cutting edge courses to ambitious individuals wanting to thrive in their field in the 21st century. We are sponsored and partnered with CAIR (Centre for Artificial Intelligence Research), MaTRIC (Centre for Research, Innovation and Coordination of Mathematics Teaching), and the Research Council of Norway.

Finally, a program dedicated to learning data science in the most intuitive project based manner. 

The Bootcamp Program

• Top Instructors and Researchers, dedicated to instructing you in the most stimulating way with video animations, quiz interactions, team building exercises
• Learn what a neural network is and how to program one from scratch, finally learn how machines actually learn
• Learn the limitations of the technology and the current possibilities so you know which business idea is possible and not possible in 2019.
• Learn how to communicate business objectives to data scientists and engineers and translate data science insights for business professionals and decision makers.
• Small class sizes ensure you have plenty of access to your instructor and can receive personalized feedback on your progress.
• Live lectures allow you to ask your instructor questions and interact with your classmates.

Watch the Video

Testimonials

Asmund Kamphaug, data scientist, now working at the police department of Oslo, stated "The best way to learn how to apply algorithms in the industry is through creating a neural network from scratch through project based learning".

The head of the Research Institution CAIR, Professor of ML Ole Christoffer Granmo, has stated: "I see great value of the diggit platform to visualize the internal dynamics of complex AI algorithms in class, as well as a problem for learners to practice building AI algorithms in their assignments."

"I think the diggitacademy system for learning is very good. It's a new way to learn new skills and I liked it. The system allows the user to get automatic feedback on their precise mistake. I am very satisfied that I can learn new technology so quickly."
Jon Magnus
ICT Consultant at Ciber Norge AS

Program Objectives

• Understand the intuition behind Artificial Neural Networks
• Apply Data preprocessing to your data utilizing python
• Understand advanced Data Architecture utilizing Python
• Understand the intuition behind Loss Function, Gradient Descent, Backpropagation through Visual Learning of Calculus
• Project: Apply Multiple features to your neural network to combine different datasets together 
• Understand the intuition behind Numerical Gradient Checking 
• Understand why gradient descent has limitations and how the industry standard BFGS algorithm solves them
• Project: Implement Overfitting, Regularization, and Testing to validate the reliability and accuracy of your Neural Network
• Apply Artificial Neural Networks in practice
• Understand the intuition behind Convolutional Neural Networks
• Be able to Apply Convolutional Neural Networks in practice
• Understand the intuition behind Recurrent Neural Networks

Course Overview

Day 1

Part 1: Recap of FeedForward Algorithm

0. Installation of Python on your local computer. 

1. Intro to Python

2. What is a variable, function, string, array

3. What is a class, constructor, why python is an object oriented language

4. Importing and cleaning Data sets

5. Pandas, Numpy and Scipy libraries

6. Data frame manipulation

7. Histograms and Probability Mass functions Notebook: Calculate and Display data

Part 2: The essentials of Linear Algebra (Finally an intuitive explanation)

1. How is a matrix defined in terms of scalars and vectors (the intuition behind linear algebra)

2. What is Numpy, what does shape mean, how do we initialize numpy arrays

3. How to visualize the transpose of a matrix

4. How do we initalize a Neural Network Class with Weights

5. What are weights and what is feedforward

Part 3: Project

1. How do we initalize a Neural Network Class with Weights

2. What are weights and what is feedforward

3. What is the sigmoid function and why do we need it

4. Project: Program our first Feedforward algorithm to predict the Bitcoin Price

5. Analyzing the results of our feedforward algorithm predicting the bitcoin price and how it compares to our prediction

Part 4: Backpropagation Algorithm

1. Recap of what we learned 

a. What is backpropagation and how does it update our weights in our neural network?

b. What is the cost function, how do we tell a Neural Network if it's wrong or right 

c. What is a derivative and why do we need the relationship between weights and cost?

d. Project: Program the derivative of the weight and cost function

e. Why do we need to derive the sigmoid function and how to do it

f. Create a cost function and know how it will be applied to our network

2. Project: Apply a training method that will update our algorithms weights

a. Learn intuitively and then implement gradient descent to train our network

3. Project: How to optimize our algorithm to predict more accurately- for instance add features (weather, day/night, news sources, health care data, timeseries data, pricing data )

Day 2

Part 1: Optimization Techniques

1. Recap of Backpropagation Algorithm 

2. Numerical Gradient Checking (an industry standard algorithm needs to perform checks to ensure incorrectly implemented gradients are identified and resolved)

3. What are the issues of training with gradient descent and how does the BFGS (Broyden-Fletcher-Goldfarb-Shanno numerical optimization algorithm) overcome the limitations of plain gradient descent by estimating the second derivative, or curvature, of the cost function surface, and using this information to make more informed movements downhill.

Part 2: Overfitting, Testing, and Regularization

1. Overfitting, Testing, and Regularization

2. So it appears our model is overfitting, but how do we know for sure?

3. How to plot the error on our training and testing sets as we train our model and identify the exact point at which overfitting begins

4. Mitigate overfitting with the technique regularization

5. How to implement regularization through adding a term to the cost function that penalizes overly complex models

6. Project: Compare our previous neural network to our evolved algorithm that provides us with more accurate training and testing errors, the ability to predict overfitting, and the ability to reduced overfitting on

Why choose this Program?

Our program utilizes the only tech learning platform that gives you automatic feedback on your precise mistake with the use of machine learning. 

Our instructors have background in education, machine learning, but also educational technology research. Meaning we combine the best of all fields to provide you a learning experience far superior to anything in the market. 

Our program is intensive, thorough, intuitive, visual, and project based. Avoid spending years in academy to master the same concepts when you can learn them in an accelerated program instead.  Your time is precious to us. When we say we will provide you with the knowledge required for the industry, we ensure that this is true our you will get your money back 100% guaranteed.

Questions?

Please contact us at communications.dpt@sci-code.com

Or call the program coordinator, Magnus Lysfjord, at +4791680446