PLSE Seminar Series - Eva Darulova - Programming with Numerical Uncertainties
Explore numerical software uncertainties in this 55-minute seminar by Eva Darulova. Delve into the challenges of finite-precision arithmetic in scientific computing and embedded systems. Learn about a programming model using real-valued specification language with error annotations, and discover how a verifying compiler selects suitable floating-point or fixed-point data types to ensure accuracy. Examine the combination of SMT theorem proving, interval and affine arithmetic, and function derivatives for precise error estimation. Investigate techniques for handling nonlinearity, discontinuities, and certain loop classes. Understand how the non-associativity of finite-precision arithmetic can be leveraged to improve accuracy through genetic programming. Gain insights into automated verification and synthesis for numerical programs, and explore the intersection of programming languages, software verification, and approximate computing.