Artificial Intelligence to solve real-world problems

There are a total of 4 questions, each with multiple parts. Please complete all the questions. Your solutions may be typed or hand-written. Please try to keep your solutions brief, succinct and to the point. You may insert additional pages to write your solutions as necessary. Please also include the name and/or web links to your reference sources.

During this course, you have hopefully gained an understanding of practical real-world use of AI-based algorithms. You have learned how AI can help transform the world around us, enabling both new technologies while improving on established ones to design intelligent systems.

In this paper, you are provided an opportunity to explore variations of methods that have been discussed in class and/or use AI to solve real-world problems in the following domains:

1) Search and Heuristics: Robotics and Games

2) Markov Decision Processes and Reinforcement Learning

3) Machine Learning and Knowledge Representation

4) Deep Learning for Natural Language Processing and Computer Vision

Please describe any algorithms that you use in the context of your selected problems and real-world scenarios.

1) Search and Heuristics – Robotics and Games (100 points)

Many approaches to AI applications involving searching in large state spaces. Thus, most AI applications and algorithms involve the use of heuristic search algorithms. Over the last several years, there has been an influx of interest in the field of heuristic based search, especially with the success of search-based planning algorithms in optimization, domain-independent planning, robotics and commercial computer games.

a) Describe a strategy and design an algorithm for finding optimal solutions to such search problems. You may use any example based on search algorithms we have studied in the class, or any combination or modified version of any heuristic based search algorithm. Please include a description of what your search space might look like, what the states might be, and what may be heuristics that could be selected.

b) What would be strengths and weaknesses of the algorithm you used in (a)?

c) Discuss whether finding an optimal solution would necessarily be needed. Describe a situation when a suboptimal solution may also be acceptable.

d) Suppose you employed a search algorithm, such as the one you described in (a), as part of a multiplayer game involving alliances among different players. Describe an example of how you would go about making optimal decisions in such a scenario.

2) Markov Decision Processes and Reinforcement Learning (100 points)

An increasing availability of rich data over recent years has led to exciting advances in the theory and practice of reinforcement learning. By learning from interactions with the environment, its goal is to find optimal or sufficiently good actions for a situation in order to maximize a reward.

a) Describe and design a real-world scenario where you can use Markov Decision Processes (MDPs).

b) Then, use Reinforcement Learning to improve the model described. Describe strengths and limitations of using reinforcement learning for your chosen problem.

You should describe the basic components of MDPs and Reinforcement Learning as they pertain to your chosen problem, define appropriate reward functions and value functions, appropriate parameters, and action selection policy. Also, discuss how you would go about modeling nondeterministic action outcomes and partial observability to maximize long terms rewards as part of your solution

3) Machine Learning and Knowledge Representation (100 points)

Knowledge representation can help to enable an intelligent machine to learn from knowledge and experience so that it can behave intelligently like a human.

Describe and design a real-world application or scenario where you would use machine learning representations to represent and process knowledge.

a) Explain your rationale behind any machine learning methodology you have selected, and how it may be effective towards addressing your real-world scenario.

b) Describe any challenges that may be encountered as well.

c) Evaluate alternate approaches you could try, and discuss why they may or may not be appropriate for representing knowledge through a machine learning system.

4) Deep Learning for Natural Language Processing and Computer Vision (100 points)

a) Natural Language Processing (50 points)

The goal of using Natural language processing (NLP) technologies is to build systems that learn to understand and represent human language and use language in appropriate context.

Describe and design a deep learning-based model which can help perform text analysis and understanding. You may use any real-life example where natural language processing is used to understand human language. Examples may include question answering, natural language inference, semantic word embeddings, and syntactic parsing.

b) Computer Vision (50 points)

Deep learning has helped bring many new applications using computer vision techniques into our daily lives.

Describe and design a deep learning-based model which can help perform computer vision analysis. You may use any real-life example where computer vision is used to understand to learning meaning from images and objects. Examples may include image classification, image transformation, object detection, segmentation, medical image analysis, etc