Algebra is one of the definitive and oldest branches of mathematics, and design of computer algorithms is one of the youngest. Despite this generation gap, the two disciplines beautifully interweave. Firstly, modern computers would be somewhat useless if they were not able to carry out arithmetic and algebraic computations efficiently, so we need to think on dedicated, sometimes rather sophisticated algorithms for these operations. Secondly, algebraic structures and theorems can help develop algorithms for things having [at first glance] nothing to do with algebra, e.g. graph algorithms.
Class Deals by MOOC List - Click here and see Coursera's Active Discounts, Deals, and Promo Codes.
One of the main goals of the offered course is thus providing the learners with the examples of the above mentioned situations. We believe the course to contain much material of interest to both CS and Math oriented students. The course is supported by programming assignments.
What You Will Learn
- Efficiently doing arithmetics on binary numbers as well as algebraic operations like polynomial multiplication, matrix multiplication and inversion.
- Design efficient algorithms problems in graph theory related to distances and matchings based on fast matrix computations and randomization.
Syllabus
WEEK 1
Arithmetics in the Realm of Circuits
In this module we will study a mathematical model of hardware: the Boolean circuits. They provide the birds eye view of how one builds a complex logical circuit from elementary building blocks by linking them with wires. We will learn efficient ways to construct circuits for all four arithmetic operations on integers represented in binary form. Among other things we will see how to efficiently calculate all carries while adding two numbers (much faster than just computing these carries sequentially) and how Newton’s approximation method for solving equations helps with implementing integer division.
WEEK 2
Boolean Circuits for Arbitrary Functions
In this second and last module devoted to Boolean circuits we move from arithmetics to the problem of constructing efficient circuits for arbitrary Boolean functions. On the way, among other things, we will see how to estimate the number of trees via DFS traversals and how to construct efficiently a circuit that computes all functions of given small number of variables.
WEEK 3
More on Multiplication of Integers and Polynomials
Multiplication of integers is among the first things people learn to do with integers at school, later moving on to higher spheres: multiplying matrices, polynomials, permutations etc. Multiplication is one of the central things in algebra. This week we focus on classical algorithms for efficient integer multiplication first, and then move on to matrix and polynomial multiplication. On the way we will see how closely integer and polynomial multiplication are really intertwined and what polynomial interpolation has to do with multiplication.
WEEK 4
Graph Reachability and Distances via Matrix Multiplication
This module is the first one to feature application pf efficient algorithms for algebraic operations to something outside algebra. Currently we turn to distances in graphs. We also study a non-typical way of multiplying matrices motivated by applications to graph reachibiilty, namely, Boolean matrix multiplication, and consider a corresponding rather general speedup technique.
WEEK 5
More on Matrix Computations
This week we first learn how to compute determinant and invert a matrix on a parallel computer, and how working with seemingly “computationally inconvenient” infinite series helps deriving quite a efficient formulas for the determinant. The second part of this module is devoted to a more common single-threaded computational model and there we prove that essentially the complexity of inverting a matrix is the same as that of matrix multiplication.
WEEK 6
Matchings in Graphs via Matrix Determinants
In the last module of the course we learn some tools for efficient use of randomization and algebra in algorithm design and use our knowledge about the existence of fast algorithms for parallel determinant computation to find perfect matchings in bipartite graphs unbelievably quickly on parallel computers.
