Dr Mehmet Suzen MInstP Theoretical Physicist | Research Scientist

• Cooperative Phenomenon: Statistical Physics Origins of Connectionist Learning
  hal-03650339 (2022)
  


• Empirical deviations of semicircle law in mixed-matrix ensembles
  hal-03464130 (2021) | notebook
  


• Equivalence in Deep Neural Networks via Conjugate Matrix Ensembles
  arXiv:2006.13687 | deep_dyson_networks.ipynb
  
  
• cascading Periodic Spectral Ergodicity (cPSE):
  A complexity measure for connectionist machine learning
  Manuscript arXiv:1911.07831
  Available in Bristol Package v0.2.6 as a single function.
  Development stage Python notebook on nbview
  Featured as Applying Occam’s razor to Deep Learning
  
  
• Generalised learning in time-series: reconstructive cross-validation (rCV)
  Manuscript arXiv:1910.09394
  Python notebook on nbview : Prototype implementation with Ornstein-Uhlenbeck processes
  Presentation February 2020
  Featured in KDnuggets
  
• Compressive transition path sampling: How to simulate rare event for longer time scales?
  Manuscript: arXiv:1804.09781
  
  
• Spectral ergodicity could explain deep learning.
  Manuscript: arXiv:1704.08303
  Software:
  • Python Package Bristol Parallel Random Matrix Tools
  Talks :
  • Spectral ergodicity for explainable deep learning:A simulation approach
  • Spectral Ergodicity in Deep Learning Architectures
  
  
• How can we test that gravity is an entropic force?
  A testable prediction from entropic gravity
  arXiv:1612.00288


• Effective ergodicity in single-spin-flip dynamics
  Phys. Rev. E 90, 032141 (2014) DOI: 10.1103/PhysRevE.90.032141
  Mehmet Suezen
  

Thesis

• Induced Charge Computation, Ph.D. Dissertation, FIAS, Goethe University, Frankfurt/Main
  Academic descendant of Carl Friedrich Gauss: math genealogy
• Chaotic behaviour of geodesics, MSc Thesis in Theoretical Physics, Cyprus

Selected and Recent Blog Posts

• Bayesian rabbit holes: Decoding conditional probability with non-commutative algebra (July 2022)
• Empirical risk minimization is not learning : A mathematical definition of learning and re-understanding of overfitting and Occam's razor in machine learning (June 2022)
• A misconception in ergodicity: Identify ergodic regime not ergodic process. (May 2022)
• Physics origins of the most important statistical ideas of recent times (January 2022)
• Deep Learning in Mind a Gentle Introduction to Spectral Ergodicity (July 2021)
• A New Matrix Mathematics for Deep Learning : Random Matrix Theory of Deep Learning (July 2021)
• Full cross-validation and generating learning curves for time-series models (June 2021)
• On the fallacy of replacing physical laws with machine learned inference systems. (April 2021)
• Statistical Physics Origins of Connectionist Learning: Cooperative Phenomenon to Ising-Lenz Architectures. (December 2020)
• A practical understanding of ergodicity (January 2020)
• Bringing back the Occam's razor to modern connectionist machine learning. (December 2019)
• On modern data scientist: A blind empiricist is not a data scientist (September 2019)
• Research as a core driver in AI and Data Science (January 2019)
• Understanding overfitting: an inaccurate meme in supervised learning (August 2017)
• Pitfalls in pseudo-random number sampling at scale with Apache Spark (July 2017)
• Practical Kullback-Leibler (KL) Divergence: Discrete Case (January 2017)
• Economy and dynamic modelling: Haavelmo's approach (July 2016)
• S-shaped data: Smoothing with quasibinomial distribution (January 2016)
• Testing emergent gravity: Gravitational Lensing to atom interferometer (January 2017)
• Understanding the empirical law of large numbers and the gambler's fallacy (August 2016)
• Is ergodicity a reasonable hypothesis? Understanding Boltzmann's ergodic hypothesis (May 2014)
• N-body wave function has no meaning but shut up and calculate: quantum bayesianism (June 2013)

Datasets

• Migrating birds real flight V-formation spatial configuration.
  zenodo (2021).
• Effective ergodicity in single-spin-flip dynamics
  zenodo (2017).

Supervisions and mentoring service to community

• Mentor Data Scientists and engineers: Machine learning and data mining projects.
• Mentor/provide guidance to academic colleagues.
• Supervise academic theses in computer science and theoretical physics.

Music

• Bruce Dickinson was kind enough to play my request on BBC

Compositions

• Othello Suite Castle I : Intro Score pdf | Midi in mp3

(c) 2016-2022 by Mehmet Suezen
suzen at acm dot org
All rights reserved
Do not copy or distribute without prior permission.
Live long and prosper in peace
In memory of Leonard Nimoy