Michael Paul Stewart Brown


Research Interests:


- Machine learning, artificial intelligence, and statistical pattern
recognition. I've been doing backprop since 1986.

- I am very fortunate to live in an era in which information is
electronically available, the volume of information is growing
exponentially, and cheap computing power allows the automated analysis
of it. This ultimately yields knowledge that enriches our lives.

- Large Language Models (LLMs) using Transformer
architectures. Interpretation of linear transform matrices and their
outliers. Efficient compute on memory limited GPUs. Relation to kernel
machines. Context size and curse of dimensionality.

- Stable Diffusion Models. Brownian motion. Stochastic differential
equations. non-ergodic processes.

- Support vector machines, boosting, mixture and product models,
information geometries, simulated annealing, Expectation Maximization,
Bayesian statistical analysis, Hidden Markov Models, stochastic
context-free grammars, Fisher features, concentration of measure tail
bounds, vector space models.

- Formal languages, algorithm development, high performance computing,
compute pipelines.

- Next Generation Sequencing, Gene Expression, Mass Spectrometry

- Characterization of information flow and mechanisms of living
things. Genomic organization, pathway inference, RNA structure,
transcription, translation, regulation, RNAi, cross species analysis.


Publications:

Papers:
Hidden Markov Models, stochastic context-free grammars, Dirichlet
mixtures, support vector machines, RNA modeling including ribosomal
RNA, RNA pseudoknot modeling, and DNA microarray data analysis....


Me:


- Email: michaelbrownid "at" gmail.com

- CV: cv.pdf

- 
  



Site Map:






  - llm_gradient_descent

  Run gradient descent on a Large Langauge Model to resuscitate
  Shakespeare.
  
  - llm_weird_basis

  Llama2. An open-science Large Languange Model that recites
  Shakespeare, talks about the climate in German, and has some weird
  basis vectors.





  - sequencing-dna-with-neural-networks
  
  Use neural networks to make noisy DNA sequencing more accurate. OK,
  how do I fit blockchain into this?

  - posterior-likelihood-prior
  
  Bayes rule illustrated. I like my distributions NORMALIZED!

  - README_tetris-gui.html
  
  Play tetris in the web browser using WebSockets treating it as
  though it were a teletype TTY printer! Modern is retro.

  - arduino-break-beam

  Measure gravity using a ardunio micro controller

  - data-science

  Opinionated take on data science boot camps (~2019)
  
  - kingsley

  A repeat expansion associated with biopolar, schizophrenia.

  - cube-rotation

  The mathematics of cube rotations. Reflecting on symmetry groups.
      




- http://www.openboundlabs.com/surf/
  Collect together some data to check the surf!




- http://www.openboundlabs.com/simplest-embedding/
  Use neural network embedding techniques in tensorflow to show you
  can learn linear orderings. Animations and Math included!

- http://www.openboundlabs.com/deepdream-flowers/
  Apply tensorflow deepdream to some images for a flower effect.



- http://www.openboundlabs.com/rare-variant-power/
  Calculate how confident you are of biological truth when you want to
  detect the presence of a minor variant but there is sequencing
  noise. Is it real or noise?

- http://www.openboundlabs.com/buckets/
  A python solution to a simple combinatorial math problem.

- http://www.openboundlabs.com/box2d-js/
  Simple physics simulation using box2d in javascript.

- http://www.openboundlabs.com/README_kez_computer.html
  General computer buying guide

- http://www.openboundlabs.com/linksFeb2017.html
  Some random links on Irreproducible Research, physics and machine
  learning, discrete math, Tensegrity.




- http://www.openboundlabs.com/bcr-abl-paperfigs/
  Results on examining minor variants in the kinase domain of the
  BCR-ABL fusion from leukemia (a drug target), time series and
  clustering plots.

- http://www.openboundlabs.com/bcrAblBarcodesFull/
  Results on examining minor variants in the kinase domain of the
  BCR-ABL fusion from leukemia (a drug target).

- http://www.openboundlabs.com/bcrabl-datatable-results/
  Results on examining minor variants in the kinase domain of the
  BCR-ABL fusion from leukemia (a drug target).

- http://www.openboundlabs.com/washubcrabl/
  Results on looking at BCR-ABL kinase domain sequencing from leukemia samples.





- http://www.openboundlabs.com/hivMI/
  Compute mutual information for all pairs of bases in HIV
  genome. Plot everything and move around it like Google maps.

- http://www.openboundlabs.com/NIAID-closehivAnalysis
  Results on resolving mixtures of full length HIV genomes (9kb) that
  differ only by a base or two

- http://www.openboundlabs.com/emory2012Feb/
  Results on examining HIV sequencing data: full HIV genome 9kb and
  GAG regions.

- http://www.openboundlabs.com/niaid/
  Results on deconvoluting mixture of 20 full HIV genomes.

- http://www.openboundlabs.com/niaid-lids/
  Results on HIV genomes in patients have Large Internal Deletions!

- http://www.openboundlabs.com/niaid-lids-validation/
  Results on HIV genomes in patients have Large Internal Deletions!

- http://www.openboundlabs.com/partner/
  Results on HIV sequences that shows 300 base deletions in ENV.

- http://www.openboundlabs.com/ucsd/
  Results on looking at HIV ENV sequencing.