Pawan > Recent Status Updates

More difficult than anticipated. The pseudo codes seemed like an actual programming language initially - expunged a lot of time to understand them. The book has its own non hardware based ISA called ILOC. IR are analogous to middlewares with optimisation runs. Many old and exotic languages are refered to, such as FORTRAN. Implementation would help with few important topics. Skipped the advanced topics.

Way too many abstractions to keep track of simultaneously. There are references to the theory of computation, data structures (particularly graphs, tables and compression), language processing, computer architecture (which I tried reading once to learn about the caches: I and the Ds). Need to refer to the newest edition.

This can wait - discounts on chocolates might be up already.

Aerodynamics is exciting because the point and line elements remind of electrostatics; vortex flow and its sheets remind of electrodynamics. Lattice (Boltzmann) method reminds of machine learning gradients, with damping factor analogous to learning rate. At higher speeds, sonic booms, wave mechanics and thermodynamics come into play - now: energy transfers in hypersonics.

History, design and applied sections make the read more interesting (and addictive).

Usually, analytical solutions have rampant approximations that extenuate drudgery. Numerically, each grid point (vertex) uses values computed in the previous step, at nearby vertices, to compute the value at current step. Large number of threads allow values to be obtained together in parallel.

Chapters 4 and 5 were of interest.

High alpha. VO2 maxed out. Boundary layer separation. Stall. Fatigue. Nose dive. Injury.

Specific part of this book was suggested for reading to understand the underlying theory in iterative mesh based simulations to select optimal design within constraints. I had read an older book long back, to appear busy, and evade speaking during discussions on blackbody radiation of fair faced females. Reading rest of it.

Professional PhD or by publication seems difficult but plausible. I used physical storage for bibliography (along with compressed cloud storage of essentials) as it is easier to run my command line tools. There was a tool called Calibre which few of us used, to organise cbz files. I have read more review papers than any other kind of paper.

Friendly read. Describes waypoint navigation. Tables (usually at the end of chapter) help recalibrate.

Enjoyed the soccer example for reduction trees.

Consecutive threads and warps minimize control divergence - wanted to try them out. The variables are in camelCase but have been using snake_case in C language - so the programs look like they contains mixed case.

Deep learning and physics follow.

A one-time affair can be accomplished by collecting all the test programs and running them on a (rented) virtual machine, chosen so as to minimize expenditure by estimating required compute capability and vetting it against hourly rates.

Instead, I ported the code to a hobby board with cortex host and 128 cores device. Test programs partition data into tiles to store them in shared memory followed by coarsening.

This book is like an extra large research paper with the appendices longer than the actual text. Rereading with suggested order:

1. Appendix A + B + C
2. Chapters 1 + 2
3. Appendix D + E + F + G
4. Chapter 3 + 4 + 5
5. Appendix H + I
6. Chapter 6
7. Appendix J + K
8. Chapter 7
9. Rest of the Appendices.

This book is like an extra large research paper with the appendices longer than the actual text. Rereading with suggested order:

1. Appendix A + B + C
2. Chapters 1 + 2
3. Appendix D + E + F + G
4. Chapter 3 + 4 + 5
5. Appendix H + I
6. Chapter 6
7. Appendix J + K
8. Chapter 7
9. Rest of the Appendices.

Not as difficult to read now. Going to try non-model based implementations in JAX, but first, remembering my kite equations using plotting libraries.

I had read a different edition of this book for a coursework until a few chapters back, which was available online for free, through none other than Google search. My mind was too shallow to estimate values for bootstrapping, back then. Time to dig.

Condensed. No conic sections. Exercises are fun - problems on a tougher side.

It is one thing to read, but another to keep track of all that is explained, lest it blows out of proportion to understanding in no time. Trees and tables are relatable; grammar from linguistics, and a dash of architecture (which needs revisiting). It would be more enjoyable to create one for a simple high level procedure or refer to another. Ich bin flummoxed.

Answering the How (to read this book):
The book is likely in print (546 pages), possibly available at bookstores. However difficult, the subject matter is pretty straightforward.

Like many others, this book too starts off with vector algebra, calculus with gradients, and then transitions to partial differential equation formulation. Boundary value problems keep coming up.

The Greek letters go all the way to vega. The finite difference method reminds of Newton's method and L-BFGS. Way too many options to keep in mind.

It is difficult to memorize all the sequences and variations - should have paid more attention to Waitzkin videos in Chessmaster. Replaying on a chessboard (as black) and against engines in chess applications (as white) is helpful.

It is interesting that a lot of it initially was farmers trying to insure and ascertain the value of their crops and produce on harvest.