Meta learning

and Evolutionary
Computing
 Team Details – G91
Sl.No Name Register Hostel/day
Number scholar
1 Ashiq Sheriff A CB.EN.U4C Day Scholar
SE21207
2 Allen Jacob George CB.EN.U4C Hosteller
SE21203
3 Navaneet N CB.EN.U4C Hosteller
SE21242
4 Nukala Venkata Durga Sandeep CB.EN.U4C Hosteller
SE21244
Evolutionary
Computation
• It is a branch of Artificial Intelligence and is used heavily for complex
optimization problems and also for continuous optimization.
• Evolutionary computation techniques are used to handle
problems that have far more variables than what traditional
algorithms can handle.
• Principles like inheritance and natural selection are used, where
the traits from the most effective models and passed on to future
generation models.
• The reason why we use evolutionary computing is because it has
the ability to produce tightly optimized solutions for a wide range
of problems.
Types of Evolutionary Algorithms

The significant ones which are currently in use;

• Genetic algorithms (GA)
• Genetic Programming (GP)
• Evolutionary Programming (EP)
• Evolutionary Strategies (ES)

Many more evolutionary algorithms also exist.

These include Gene Expression Programming,
Differential Evolution, Learning Classifier
Systems and Neuro-evolution.
Genetic algorithms (GA)

• This algorithm is used to solve optimization and search

problems by mimicking the process of natural selection.
• It works by evolving a population of candidate
solutions(the potential solutions to the problem being
addressed) over several generations to find the best or
near-best solution to a problem.
• There are some key concepts and algorithms which we
would see below;
Population, Selection, Crossover and Mutation
Key concepts of Genetic algorithm:
Population: A set of candidate solutions to the problem being
addressed.
Selection: The process of choosing individuals from the
current population to create offspring for the next generation.
Crossover: The process of combining parts of two parent
chromosomes to produce one or more offspring.
Mutation: The process of randomly altering one or more
genes in a chromosome.
Note:
Chromosome is the encoded versions of candidate solutions,
typically represented as strings (binary, real-valued, etc.).
Genetic Programming (GP)

• It is a type of evolutionary algorithm and a subset of

genetic algorithms used in evolutionary computing.
• It evolves computer programs to solve problems by
mimicking the process of natural selection.
• It also evolves tree structures representing hierarchical
computer programs.

There are many key features as far as GP is concerned,

out of which 3 important features would be
discussed below:
Key features of Genetic programming

• Tree-Based Representation: Candidate solutions are

typically represented as tree structures.
• Evolution of Programs: It evolves entire programs to find
the best solution to a problem.
• Fitness Function: It evaluates how well a candidate
program solves the problem at hand.

Other features of GP:

Genetic Operators, Selection, Initialization, Termination
Criteria.
Evolutionary Programming (EP)

• In Evolutionary Programming, the programs that need

to be optimized have a fixed structure, while the
numeric parameters can evolve.
• The main operator of evolutionary programming is
mutation.
• In evolutionary programming, members of the
population are seen as part of a specific species rather
than members of the same species. Every parent
generates an offspring by using a (μ + μ) survivor
selection.
Key features of EP

• Emphasis on Mutation: Evolutionary programming

primarily uses mutation as its genetic operator, rather
than crossover.
• Population-Based Search: EP maintains a population of
candidate solutions that evolve over time.
• Simulation of Evolutionary Processes: EP simulates the
process of natural evolution to evolve solutions.
Evolutionary Strategies (ES)

• Evolutionary strategies usually work by making use of

self-adaptive mutation rates.
• They work with vectors of real numbers as
representations of solutions.
• Evolutionary strategies are optimization techniques that
are based on the ideas of evolution.
• They use natural problem-dependent representations
and mainly make use of mutation and selection as
search operators.
Key features of Evolutionary strategies

• Self-Adaptive Mutation: Evolutionary strategies often use

self-adaptive mutation mechanisms to adjust the
mutation rate and variance during the optimization
process.
• Recombination and Selection: Evolutionary strategies
typically use recombination (or crossover) and selection
to evolve candidate solutions.
• Continuous Optimization: Evolutionary strategies are
particularly suited for continuous optimization problems,
where the search space consists of continuous variables
rather than discrete ones.
Meta learning algorithms

• ES-MAML:
Integrates Evolution Strategies (ES) with Model-Agnostic Meta-Learning (MAML).
ES is used to optimize the meta-parameters of the learning
algorithm.

• GA-HPO:
Utilizes Genetic Algorithms (GAs) to optimize hyperparameters for meta-learning
models.

• Neuro-evolution for Neural Architecture Search

(NAS):
Uses evolutionary algorithms to search for optimal neural network architectures.
Example algorithms: Neuro-evolution of Augmenting Topologies (NEAT) ,
Algorithmic Trading with Meta-
Learning
Develop a meta-learning model for algorithmic trading that
adapts to changing market conditions and learns optimal
trading strategies over time.
Approach :
• Collect historical trading data from different market
conditions.
• Use a meta-learning algorithm like MAML to train a model that
can quickly adapt to new market conditions.
• Implement a reinforcement learning component to
continuously improve the trading strategy based on real-time
market feedback.
Meta-Learning for Medical Image
Analysis
Create a meta-learning-based system for classifying medical
images and detecting abnormalities with few-shot learning
capabilities.
Approach:
• Gather a diverse set of labeled medical images (e.g., X-rays,
MRIs, CT scans).
• Use Prototypical Networks or Siamese Networks to develop a
model that can classify new medical images with limited
labeled examples.
• Continuously update the model to incorporate new imaging
data and improve accuracy.
Meta-Learning for Intelligent
Assessment Systems
Develop an intelligent assessment system that adapts to the
student's knowledge level and provides personalized feedback.
Approach:
• Gather data from students' interactions with assessment tools,
including answers, response times, and patterns of errors.
• Use a meta-learning algorithm to train a model that can adapt
assessment difficulty and feedback based on student
performance.
• Implement continuous updates to the model as new data
becomes available.