From 84322785a9fc8cbbacac43ef45b15325cda13875 Mon Sep 17 00:00:00 2001 From: Noah Smith Date: Sun, 2 Jun 2024 11:56:27 +0200 Subject: [PATCH] nineties --- _drafts/nineties.md | 7 ++++--- _posts/2024-04-07-unix.md | 1 - _posts/2024-04-13-pdp-10.md | 1 - _posts/2024-04-16-aerospace-numerical-computing.md | 1 - _posts/2024-04-17-simplex-algorithm.md | 1 - _posts/2024-04-18-kalman-filter.md | 1 - _posts/2024-04-20-modern-kalman-filter.md | 1 - _posts/2024-04-21-particle-filter.md | 1 - _posts/2024-04-28-vax.md | 1 - _posts/2024-05-05-sun.md | 1 - _posts/2024-05-06-vms.md | 1 - _posts/2024-05-14-lost-in-space.md | 1 - _posts/2024-05-17-starid77.md | 1 - _posts/2024-05-20-starid93-triangles.md | 1 - _posts/2024-05-24-starid97a-grid-algorithm.md | 1 - _posts/2024-05-26-starid86-hubble.md | 1 - _posts/2024-05-31-starid89-sirtf.md | 1 - _posts/2024-06-03-nineties.md | 13 +++++++++++++ 18 files changed, 17 insertions(+), 19 deletions(-) create mode 100644 _posts/2024-06-03-nineties.md diff --git a/_drafts/nineties.md b/_drafts/nineties.md index 2ee38ee..e7e30bb 100644 --- a/_drafts/nineties.md +++ b/_drafts/nineties.md @@ -1,12 +1,13 @@ --- layout: post title: "nineties" -categories: aerospace starid --- [paper 1997b](https://statespace.dev/docs/papers/1997%20lindblad.pdf){:target="_blank" rel="noopener"} -this second paper from 1997 is an interesting case, as it's the first in this collection to bring in the topic of neural networks. it also summarizes three similar papers, but they're pay-walled. a discussion of wider context seems useful at this point to set the stage for a separate detailed discussion. +this paper is an interesting case, as it's the first to bring in the topic of neural networks. it mentions three previous related papers, but they're pay-walled. in an [earlier post](https://statespace.dev/starid77.html), mention is made of the resonance between the fortran era and the cold war, from 57 to 77. this paper is part of the final stages of that era, twenty years later. -in an earlier discussion of a 1977 paper, mention is made of the resonance from 57 to 77 of the [fortran era and peak cold war](https://statespace.dev/aerospace/starid/starid77.html). there's an additional periodicity here, as this 1997 paper provides a time-capsule of the cold war's end. for those involved with astronomy and aerospace engineering around rlm and wrw at utaustin in 1990, the world had changed completely. aerospace shrank dramatically, as did neural networks. after another twenty years, both revived in an equally dramatic fashion. this is an interesting part of the project starid story to gradually explore. +for those involved with astronomy and aerospace engineering in the nineties, around rlm and wrw at utaustin for example, the world changed. aerospace shrank dramatically, as did neural networks. thirty years later, nearing 2027, both have revived in an equally surprising fashion. all this is an interesting aspect of the project starid story, to be explored gradually over time. for an overview of neural networks [the brain makers](https://a.co/d/6QQ1ggo) is recommendable, and made for interesting reading around 2010 while relaxing 'at the office' in the [mcc building](https://en.wikipedia.org/wiki/Microelectronics_and_Computer_Technology_Corporation), since mcc plays a prominent role within the book. +a notable thing about cold war era neural networks is that they generally involved alternative 'parallel' hardware schemes, in direct opposition to classic, sequential, 'von neumann' hardware. simd 'single instruction multiple data' hardware was downplayed as a compromise, and it's interesting to consider how simd has in fact become completely dominant. one question might be how many connectionists understood and used linear algebra as an inherent part of their field, given that so many of them came from outside the 'hard' physical fields? this paper discusses using commercial ibm hardware and serves as a time-capsule from the brief ascendancy of neural network chips and lisp machines. +one motivation is speed. parallel hardware neurons within the chip simultaneously process a star tracker image into a connectionist classification. in some sense starid becomes a single 'wide operation', as opposed to a long sequence of 'narrow operations' with classic hardware. another motivation is robustness and 'fuzzy' resistance to noise. the processing performed by the neurons essentially checks the dot products of pairs of vectors. a match can be 'close enough', there's built-in tolerance to variations. diff --git a/_posts/2024-04-07-unix.md b/_posts/2024-04-07-unix.md index 7ffdcda..3c914b9 100644 --- a/_posts/2024-04-07-unix.md +++ b/_posts/2024-04-07-unix.md @@ -1,7 +1,6 @@ --- layout: post title: unix -#categories: computing unix --- [photos](https://photos.app.goo.gl/rL5NTL2iFomFjedM6){:target="_blank" rel="noopener"} diff --git a/_posts/2024-04-13-pdp-10.md b/_posts/2024-04-13-pdp-10.md index 0956931..58a3418 100644 --- a/_posts/2024-04-13-pdp-10.md +++ b/_posts/2024-04-13-pdp-10.md @@ -1,7 +1,6 @@ --- layout: post title: pdp-10 -#categories: computing pdp-10 --- [photos](https://photos.app.goo.gl/rL5NTL2iFomFjedM6){:target="_blank" rel="noopener"} diff --git a/_posts/2024-04-16-aerospace-numerical-computing.md b/_posts/2024-04-16-aerospace-numerical-computing.md index 5f4cb85..506f586 100644 --- a/_posts/2024-04-16-aerospace-numerical-computing.md +++ b/_posts/2024-04-16-aerospace-numerical-computing.md @@ -1,7 +1,6 @@ --- layout: post title: "aerospace numerical computing" -#categories: aerospace numerical --- this is an initial experiment with bringing some legacy 'blog style discussion' from python sphinx forward into website jekyll. diff --git a/_posts/2024-04-17-simplex-algorithm.md b/_posts/2024-04-17-simplex-algorithm.md index 77da281..d5d55f4 100644 --- a/_posts/2024-04-17-simplex-algorithm.md +++ b/_posts/2024-04-17-simplex-algorithm.md @@ -1,7 +1,6 @@ --- layout: post title: "simplex algorithm" -#categories: aerospace simplex --- in the early days of digital computing, what were some of the motivations for industry to invest in computers? what exactly could computers do? they could perform numerical simulations. in the manhattan project, they replaced rooms of humans operating mechanical calculators. and they could collate data. in the sage continental air defence system, they replaced humans in low level command and control roles. diff --git a/_posts/2024-04-18-kalman-filter.md b/_posts/2024-04-18-kalman-filter.md index c2789d2..0b4c6a7 100644 --- a/_posts/2024-04-18-kalman-filter.md +++ b/_posts/2024-04-18-kalman-filter.md @@ -1,7 +1,6 @@ --- layout: post title: "kalman filter" -#categories: aerospace kalman --- as both intercontinental missiles and thermonuclear weapons became practical in the mid fifties, one more technology became essential - self-contained guidance systems that could reliably get near enough to a spot on the other side of the globe. what was needed was an autonomous digital system for guidance and control of long range missiles. this was an important motivation behind the appearance, at the beginning of the sixties, of the kalman filter. diff --git a/_posts/2024-04-20-modern-kalman-filter.md b/_posts/2024-04-20-modern-kalman-filter.md index 0ca3322..ad4802e 100644 --- a/_posts/2024-04-20-modern-kalman-filter.md +++ b/_posts/2024-04-20-modern-kalman-filter.md @@ -1,7 +1,6 @@ --- layout: post title: "modern kalman filter" -#categories: aerospace kalman --- the modern kalman filter is an intermediate step between the classical kalman filter, where uncertainty is represented as gaussian, and the particle filter. the advantage of gaussians is that they have only two parameters, so asked the bare minimum from late fifties computing hardware. much effort went into efficiently evolving the values forward through time via the matrix riccati differential equation for the covariance. diff --git a/_posts/2024-04-21-particle-filter.md b/_posts/2024-04-21-particle-filter.md index b7459a3..17a4cd9 100644 --- a/_posts/2024-04-21-particle-filter.md +++ b/_posts/2024-04-21-particle-filter.md @@ -1,7 +1,6 @@ --- layout: post title: "particle filter" -#categories: aerospace particle --- from the modern kalman filter, the next step forward with 'sample-and-propagate' algorithms was to use non-deterministic sampling points or particles. a descriptive name for this is sequential monte carlo sampling processor, but a common short-hand is particle filter. diff --git a/_posts/2024-04-28-vax.md b/_posts/2024-04-28-vax.md index aac14f9..62c1642 100644 --- a/_posts/2024-04-28-vax.md +++ b/_posts/2024-04-28-vax.md @@ -1,7 +1,6 @@ --- layout: post title: "vax-11/780" -#categories: computing vax --- [photos](https://photos.app.goo.gl/rL5NTL2iFomFjedM6){:target="_blank" rel="noopener"} diff --git a/_posts/2024-05-05-sun.md b/_posts/2024-05-05-sun.md index c79005d..e6654b6 100644 --- a/_posts/2024-05-05-sun.md +++ b/_posts/2024-05-05-sun.md @@ -1,7 +1,6 @@ --- layout: post title: "sun-1/150" -#categories: computing sun --- [photos](https://photos.app.goo.gl/rL5NTL2iFomFjedM6){:target="_blank" rel="noopener"} diff --git a/_posts/2024-05-06-vms.md b/_posts/2024-05-06-vms.md index e45744c..62b2d1b 100644 --- a/_posts/2024-05-06-vms.md +++ b/_posts/2024-05-06-vms.md @@ -1,7 +1,6 @@ --- layout: post title: "vms" -#categories: computing vms --- [photos](https://photos.app.goo.gl/rL5NTL2iFomFjedM6){:target="_blank" rel="noopener"} diff --git a/_posts/2024-05-14-lost-in-space.md b/_posts/2024-05-14-lost-in-space.md index 93a5bd4..39dd769 100644 --- a/_posts/2024-05-14-lost-in-space.md +++ b/_posts/2024-05-14-lost-in-space.md @@ -1,7 +1,6 @@ --- layout: post title: "lost in space problem" -#categories: aerospace starid --- the core of the lost in space problem is a total lack of star tracker pointing knowledge. often there's at least some knowledge of the star tracker's attitude, but the lost in space problem is the extreme opposite case, with infinite attitude uncertainty. diff --git a/_posts/2024-05-17-starid77.md b/_posts/2024-05-17-starid77.md index 9aaa04d..bafca81 100644 --- a/_posts/2024-05-17-starid77.md +++ b/_posts/2024-05-17-starid77.md @@ -1,7 +1,6 @@ --- layout: post title: "starid77" -#categories: aerospace starid --- [paper 1977](https://statespace.dev/docs/papers/1977%20junkins.pdf){:target="_blank" rel="noopener"} diff --git a/_posts/2024-05-20-starid93-triangles.md b/_posts/2024-05-20-starid93-triangles.md index 6aa9f49..687b2b8 100644 --- a/_posts/2024-05-20-starid93-triangles.md +++ b/_posts/2024-05-20-starid93-triangles.md @@ -1,7 +1,6 @@ --- layout: post title: "starid93 triangles" -#categories: aerospace starid --- [paper 1993](https://statespace.dev/docs/papers/1993%20liebe.pdf){:target="_blank" rel="noopener"} diff --git a/_posts/2024-05-24-starid97a-grid-algorithm.md b/_posts/2024-05-24-starid97a-grid-algorithm.md index d50e160..4f398df 100644 --- a/_posts/2024-05-24-starid97a-grid-algorithm.md +++ b/_posts/2024-05-24-starid97a-grid-algorithm.md @@ -1,7 +1,6 @@ --- layout: post title: "starid97a grid algorithm" -#categories: aerospace starid --- [paper 1997a](https://statespace.dev/docs/papers/1997%20padgett.pdf){:target="_blank" rel="noopener"} diff --git a/_posts/2024-05-26-starid86-hubble.md b/_posts/2024-05-26-starid86-hubble.md index 77f188e..9f1008c 100644 --- a/_posts/2024-05-26-starid86-hubble.md +++ b/_posts/2024-05-26-starid86-hubble.md @@ -1,7 +1,6 @@ --- layout: post title: "starid86 hubble" -#categories: aerospace starid --- [paper 1986](https://statespace.dev/docs/papers/1986%20groth.pdf){:target="_blank" rel="noopener"} diff --git a/_posts/2024-05-31-starid89-sirtf.md b/_posts/2024-05-31-starid89-sirtf.md index d04abba..215ae5f 100644 --- a/_posts/2024-05-31-starid89-sirtf.md +++ b/_posts/2024-05-31-starid89-sirtf.md @@ -1,7 +1,6 @@ --- layout: post title: "starid89 sirtf" -#categories: aerospace starid --- [paper 1989](https://statespace.dev/docs/papers/1989%20bezooijen.pdf){:target="_blank" rel="noopener"} diff --git a/_posts/2024-06-03-nineties.md b/_posts/2024-06-03-nineties.md new file mode 100644 index 0000000..849564e --- /dev/null +++ b/_posts/2024-06-03-nineties.md @@ -0,0 +1,13 @@ +--- +layout: post +title: "nineties" +--- +[paper 1997b](https://statespace.dev/docs/papers/1997%20lindblad.pdf){:target="_blank" rel="noopener"} + +this paper is an interesting case, as it's the first to bring in the topic of neural networks. it mentions three previous related papers, but they're pay-walled. in an [earlier post](https://statespace.dev/starid77.html), mention is made of the resonance between the fortran era and the cold war, from 57 to 77. this paper is part of the final stages of that era, twenty years later. + +for those involved with astronomy and aerospace engineering in the nineties, around rlm and wrw at utaustin for example, the world changed. aerospace shrank dramatically, as did neural networks. thirty years later, nearing 2027, both have revived in an equally surprising fashion. all this is an interesting aspect of the project starid story, to be explored gradually over time. for an overview of neural networks [the brain makers](https://a.co/d/6QQ1ggo) is recommendable, and made for interesting reading around 2010 while relaxing 'at the office' in the [mcc building](https://en.wikipedia.org/wiki/Microelectronics_and_Computer_Technology_Corporation), since mcc plays a prominent role within the book. + +a notable thing about cold war era neural networks is that they generally involved alternative 'parallel' hardware schemes, in direct opposition to classic, sequential, 'von neumann' hardware. simd 'single instruction multiple data' hardware was downplayed as a compromise, and it's interesting to consider how simd has in fact become completely dominant. one question might be how many connectionists understood and used linear algebra as an inherent part of their field, given that so many of them came from outside the 'hard' physical fields? this paper discusses using commercial ibm hardware and serves as a time-capsule from the brief ascendancy of neural network chips and lisp machines. + +one motivation was speed. parallel hardware neurons within the chip simultaneously process a star tracker image into a connectionist classification. in some sense starid becomes a single 'wide operation', as opposed to a long sequence of 'narrow operations' with classic hardware. another motivation was robustness and 'fuzzy' resistance to noise. the processing performed by the neurons essentially checks the dot products of pairs of vectors. a match can be 'close enough', there's a built in tolerance to variations.