Pinned Repositories
Coinz
Frederik Kelbel s1653529
CV_Website
GAN-RNN_Experiments
Various experiments with Generative Adverserial and Recurrent Neural Networks
GPandGA
GP/GA-algorithm for prime-number generation
HitNsplit
Android game made with Unity
IAML2018
Introductory Applied Machine Learning [Labs & CourseWork]
IVR-assignment
MLP_CW4
Optimal-Control-of-Agent-Based-Dynamics
Ever since the concepts of dynamic programming wereintroduced, one of the most difficult challenges has been to adequatelyaddress high-dimensional control problems. With growing dimension-ality, the utilisation of Deep Neural Networks promises to circumventthe issue of an otherwise exponentially increasing complexity. Thepaper specifically investigates the sampling issues the Deep GalerkinMethod is subjected to. It proposes a drift relaxation-based samplingapproach to alleviate the symptoms of high-variance policy approxi-mations. This is validated on mean-field control problems; namely, thevariations of the opinion dynamics presented by the Sznajd and theHegselmann-Krause model. The resulting policies induce a significantcost reduction over manually optimised control functions and show improvements on the Linear-Quadratic Regulator problem over the Deep FBSDE approach.
TradingBot
Selenium based trading agent for Trading212
FreditorK's Repositories
FreditorK/Optimal-Control-of-Agent-Based-Dynamics
Ever since the concepts of dynamic programming wereintroduced, one of the most difficult challenges has been to adequatelyaddress high-dimensional control problems. With growing dimension-ality, the utilisation of Deep Neural Networks promises to circumventthe issue of an otherwise exponentially increasing complexity. Thepaper specifically investigates the sampling issues the Deep GalerkinMethod is subjected to. It proposes a drift relaxation-based samplingapproach to alleviate the symptoms of high-variance policy approxi-mations. This is validated on mean-field control problems; namely, thevariations of the opinion dynamics presented by the Sznajd and theHegselmann-Krause model. The resulting policies induce a significantcost reduction over manually optimised control functions and show improvements on the Linear-Quadratic Regulator problem over the Deep FBSDE approach.
FreditorK/HitNsplit
Android game made with Unity
FreditorK/Coinz
Frederik Kelbel s1653529
FreditorK/CV_Website
FreditorK/GAN-RNN_Experiments
Various experiments with Generative Adverserial and Recurrent Neural Networks
FreditorK/GPandGA
GP/GA-algorithm for prime-number generation
FreditorK/IAML2018
Introductory Applied Machine Learning [Labs & CourseWork]
FreditorK/IVR-assignment
FreditorK/MLP_CW4
FreditorK/TradingBot
Selenium based trading agent for Trading212
FreditorK/Quantum_Comp_Experiments
Learning various things with Q#.