5.8 Shapley Values | Interpretable Machine Learning

9 days ago by pozorvlak

The Shapley value of a feature value is not the difference of the predicted value after removing the feature from the model training. The interpretation of the Shapley value is: Given the current set of feature values, the contribution of a feature value to the difference between the actual prediction and the mean prediction is the estimated Shapley value.

The Shapley value is the wrong explanation method if you seek sparse explanations (explanations that contain few features). Explanations created with the Shapley value method always use all the features. Humans prefer selective explanations, such as those produced by LIME. LIME might be the better choice for explanations lay-persons have to deal with. Another solution is SHAP introduced by Lundberg and Lee (2016)41, which is based on the Shapley value, but can also provide explanations with few features.

computers
programming
machinelearning
maths
gametheory
statistics
The Shapley value is the wrong explanation method if you seek sparse explanations (explanations that contain few features). Explanations created with the Shapley value method always use all the features. Humans prefer selective explanations, such as those produced by LIME. LIME might be the better choice for explanations lay-persons have to deal with. Another solution is SHAP introduced by Lundberg and Lee (2016)41, which is based on the Shapley value, but can also provide explanations with few features.

9 days ago by pozorvlak

Structured disentangled representations

19 days ago by pozorvlak

How to interpret the various terms in the ELBO objective function, and how to generalise it to variational autoencoders (VAEs) with several latent variables.

computers
programming
maths
statistics
machinelearning
ai
deeplearning
19 days ago by pozorvlak

Why are 2D vector graphics so much harder than 3D? | Clean Rinse

25 days ago by pozorvlak

Featuring a surprise appearance by the Abel-Ruffini theorem!

computers
graphics
3dmodelling
fonts
tex
postscript
maths
25 days ago by pozorvlak

Don’t waste your time on statistics - Towards Data Science

25 days ago by pozorvlak

So if you’re dealing with uncertainty (e.g. “Will this machine learning system work on tomorrow’s data?”) and the options aren’t each alike in dignity (e.g. “We probably shouldn’t launch it unless it works.”) then you’ve come to the right place: statistics is for you. Zoom through its main ideas here. Everyone else, flee now before you end up crunching a bunch of numbers meticulously… and uselessly. Analytics is a better option for you.

maths
statistics
datascience
25 days ago by pozorvlak

The Behavioral Approach to Open and Interconnected Systems

5 weeks ago by pozorvlak

Studying control theory via relations rather than functions, because real systems have feedback loops rather than strict inputs and outputs.

maths
physics
engineering
controltheory
5 weeks ago by pozorvlak

Common probability misunderstanding: half heads?

7 weeks ago by pozorvlak

As n -> infinity, the probability of getting exactly n/2 heads tends to zero.

maths
statistics
probability
7 weeks ago by pozorvlak

Golden strings & rabbit constant

maths

7 weeks ago by pozorvlak

Golden strings are analogous to Fibonacci numbers, except one uses concatenation rather than addition. Start with s1 = "1" and s2 = "10". Then define sn = sn-1

7 weeks ago by pozorvlak

Floating point error is the least of my worries

maths
modelling
computers
programming
floatingpoint

8 weeks ago by pozorvlak

Modeling error is usually several orders of magnitude greater than floating point error. People who nonchalantly model the real world and then sneer at floating point as just an approximation strain at gnats and swallow camels.

8 weeks ago by pozorvlak

SafeCurves: Introduction

security
infosec
crypto
maths
computers

8 weeks ago by pozorvlak

Secure implementations of the standard curves are theoretically possible but very hard.

Most of these attacks would have been ruled out by better choices of curves that allow simple implementations to be secure implementations. This is the primary motivation for SafeCurves. The SafeCurves criteria are designed to ensure ECC security, not just ECDLP security.

8 weeks ago by pozorvlak

Curve Ed448 and Karatsuba multiplication

crypto
maths
algorithms
computers
security
infosec

8 weeks ago by pozorvlak

The elliptic curve Ed448 is nicknamed "Goldilocks" for reasons explained here. It uses an algorithm requiring three multiplications where you'd expect four.

8 weeks ago by pozorvlak

What leads to success at math contests? | Power Overwhelming

9 weeks ago by pozorvlak

"What leads to success at math contests?"

Highly generalisable advice, related somewhat to…

maths
learning
olympiads
Highly generalisable advice, related somewhat to…

9 weeks ago by pozorvlak

Why Do We Pay Pure Mathematicians? – Math with Bad Drawings

9 weeks ago by pozorvlak

"Like most researchers in her subfield [she] considers any number larger than 5 to be monstrously big."

maths
comics
9 weeks ago by pozorvlak

Napkin | Power Overwhelming

9 weeks ago by pozorvlak

Huge amounts of maths explained as if on the back of a napkin.

maths
9 weeks ago by pozorvlak

Against the “Research vs. Olympiads” Mantra | Power Overwhelming

9 weeks ago by pozorvlak

But we need this kind of problem-solving skill and talent too much for it to all be spent on computing R(6,6).

maths
olympiads
research
jobs
9 weeks ago by pozorvlak

foxes are totally trustworthy — naamahdarling: thetasteoffire: ...

9 weeks ago by pozorvlak

Just as the phrase “what the entire fuck” implies the existence of fractional fucks, the phrase “what the absolute fuck” implies the existence of both positive and negative fucks (or else there would be no need for an absolute value operation). Taken together with the phrase “what the actual fuck” (which implies the existence of imaginary fucks), we may thus conclude that fuckery is isomorphic with the complex field.

maths
funny
9 weeks ago by pozorvlak

Following an idea to its logical conclusion

models
maths
measurement
geometry
physics

10 weeks ago by pozorvlak

Following an idea to its logical conclusion might be extrapolating a model beyond its valid range.

10 weeks ago by pozorvlak

Going Critical — Melting Asphalt

may 2019 by pozorvlak

What I learned from the simulation above is that there are ideas and cultural practices that can take root and spread in a city that simply can't spread out in the countryside. (Mathematically can't.) These are the very same ideas and the very same kinds of people. It's not that rural folks are e.g. "small-minded"; when exposed to one of these ideas, they're exactly as likely to adopt it as someone in the city. Rather, it's that the idea itself can't go viral in the countryside because there are...

networking
socialnetworks
publichealth
vaccines
maths
dataviz
science
badscience
academia
may 2019 by pozorvlak

realhats v3.0

april 2019 by pozorvlak

realhats is a package for LATEX that makes the \hat command put real hats on symbols.

tex
maths
funny
unicode
april 2019 by pozorvlak

Good–Turing frequency estimation - Wikipedia

april 2019 by pozorvlak

Good–Turing frequency estimation is a statistical technique for estimating the probability of encountering an object of a hitherto unseen species, given a set of past observations of objects from different species. In drawing balls from an urn, the 'objects' would be balls and the 'species' would be the distinct colors of the balls (finite but unknown in number). After drawing R red {\displaystyle R_{\text{red}}} R_\text{red} red balls, R black {\displaystyle R_{\text{black}}} R_\text{black} black balls and R green {\displaystyle R_{\text{green}}} R_\text{green} green balls, we would ask what is the probability of drawing a red ball, a black ball, a green ball or one of a previously unseen color.

maths
statistics
algorithms
april 2019 by pozorvlak

Mean, variance, skewness & kurtosis computed with a fold

maths
programming
statistics
algorithms
haskell
probability

april 2019 by pozorvlak

The sample statistics mean, variance, skewness, and kurtosis can be calculated in one pass through the data. We show how to do this with a fold operation.

april 2019 by pozorvlak

Runge-Kutta implemented as a fold in Haskell

computers
programming
maths
algorithms
haskell

april 2019 by pozorvlak

An ordinary differential equation solver like the famous Runge Kutta method is naturally expressed as a fold in functional programming.

april 2019 by pozorvlak

Two types of viewpoint | David R. MacIver

march 2019 by pozorvlak

A thamagar viewpoint can’t see the wood for the trees. A relip one can’t see the leaves for the tree. When you work in your area of expertise, your viewpoint is thamagar. When you explain it to someone unfamiliar with it, you try to give them a relip view.

The way you look at the ingroup is usually thamagar, and the way you look at the outgroup is usually relip. Combinatorics is thamagar, category theory is relip. Writing a poem requires you to be thamagar, teaching someone requires you to be relip.

teaching
maths
learning
The way you look at the ingroup is usually thamagar, and the way you look at the outgroup is usually relip. Combinatorics is thamagar, category theory is relip. Writing a poem requires you to be thamagar, teaching someone requires you to be relip.

march 2019 by pozorvlak

kenbot on Twitter: "My best kept secret* is that string & wiring diagrams--plucked straight out of applied category theory--are *fabulous* for software and system design. And you don't have to learn a damn thing to start using them. 🙌 *will talk to

twitter maths categorytheory programming softwarearchitecture

march 2019 by pozorvlak

twitter maths categorytheory programming softwarearchitecture

march 2019 by pozorvlak

The woman who invented abstract algebra | Cosmos

maths
antisemitism
sexism
physics
nazism
womeninstem
women

march 2019 by pozorvlak

In Noether’s time, the scientific establishment worked hard to keep women out. A genius of Noether’s calibre, with Einstein’s backing, could maybe be included. Even today, in mathematics or physics, we can observe an asymmetry in the treatment of women and men in academia.

And as Emmy Noether taught us, whenever a symmetry is broken, that means something is being lost.

march 2019 by pozorvlak

Going beyond the Golden Ratio. | Extreme Learning

march 2019 by pozorvlak

The three "most irrational" (hardest to approximate with rationals) numbers are

- phi = (1 + sqrt(5))/2, the Golden Ratio

- 1 + sqrt(2)

- (9 + sqrt(221))/10

maths
dataviz
- phi = (1 + sqrt(5))/2, the Golden Ratio

- 1 + sqrt(2)

- (9 + sqrt(221))/10

march 2019 by pozorvlak

The Math That Unifies the Laws of Physics

maths
physics
quantum

march 2019 by pozorvlak

In my 50s, too old to become a real expert, I have finally fallen in love with algebraic geometry. As the name suggests, this is the…

march 2019 by pozorvlak

Unscrambling the Hidden Secrets of Superpermutations | Quanta Magazine

maths
sf
superpermutations
computers
algorithms
4chan

february 2019 by pozorvlak

A science fiction novelist and an internet commenter made breakthroughs on a longstanding problem about the number of ways you can arrange a set of items. What

february 2019 by pozorvlak

Open Letter: An Open Letter to the Mathematical Community - McSweeney’s Internet Tendency

january 2019 by pozorvlak

We should return to our Pythagorean roots, and become a cult.

maths
cults
funny
university
teaching
january 2019 by pozorvlak

[1806.07366] Neural Ordinary Differential Equations

january 2019 by pozorvlak

We introduce a new family of deep neural network models. Instead of specifying a discrete sequence of hidden layers, we parameterize the derivative of the hidden state using a neural network. The output of the network is computed using a black-box differential equation solver. These continuous-depth models have constant memory cost, adapt their evaluation strategy to each input, and can explicitly trade numerical precision for speed. We demonstrate these properties in continuous-depth residual networks and continuous-time latent variable models. We also construct continuous normalizing flows, a generative model that can train by maximum likelihood, without partitioning or ordering the data dimensions. For training, we show how to scalably backpropagate through any ODE solver, without access to its internal operations. This allows end-to-end training of ODEs within larger models.

maths
deeplearning
machinelearning
computers
programming
ai
january 2019 by pozorvlak

The FedEx Problem | Hacker News

january 2019 by pozorvlak

How did FedEx choose their hub airport? How do they do flight scheduling? Some war stories from an early-hire operational researcher.

aircraft
shippingcontainers
optimization
startups
business
maths
january 2019 by pozorvlak

Unprovability comes to machine learning

january 2019 by pozorvlak

Ben-David and colleagues then prove that the ability to carry out a weak form of monotone compression is related to the size of certain infinite sets. The set that the authors ultimately use in their work is the unit interval, which is the set of real numbers between 0 and 1. Their results imply that the finite subsets of the unit interval have monotone-compression schemes, and therefore are learnable in EMX, if and only if the continuum hypothesis is true, which is known to be unprovable.

machinelearning
maths
logic
january 2019 by pozorvlak

Copy this bookmark: