Default page

This default page appears to be a statistician’s blog. Hopefully this clarifies some things.

How much calories does an LLM burn?

llm

deep learning

ai

energy

A whimsical attempt at making sense of LLM energy consumption

Thinking is not hierarchical, why should AI be?

machine learning

deep learning

artificial intelligence

A comment on Yann LeCun’s A Path Towards Autonomous AI

Word embedding similarity is tricky

python

rust

word embeddings

machine learning

Even though word embeddings are widely used, we do not know a whole lot about them

Building a property graph database from scratch

python

neo4j

graph

database

Graph databases have gained in popularity. They rely on simply but effective building blocks, which this post will discuss.

Mapping spans from one document to another, an excellent interview question

leadership

interviews

python

Designing suitable questions for technical interviews is tricky. Using a real-world example, we’ll discuss what can set your questions apart.

The relationship between L2 norm and dropout in linear regression

python

deep learning

L2 norm and dropout are widely regarded to be equivalent in regression models, which this blog post will challenge

Playing “The game” with Monte Carlo simulations

games

python

simulation

Sometimes playing a game is not enough, and one wants to figure out how to beat it.

Apple may be the last stronghold for privacy

apple

privacy

In an interesting turn of events Apple will inadvertently become the leading company to protect consumer privacy

NVIDIA resolves gRPC performance issues

machine learning

nvidia

triton inference server

Using Triton Inference Server’s gRPC interface one could encounter massive performance drops. Taking a look at the benchmark shows the regression has been resolved!

Graph randomization is not the same as graph randomization

network science

jupyter

In network science, node degree and strength are important properties. When randomising networks one needs to take them into consideration

The missing spaCy benchmark

python

spacy

nlp

spaCy is a self-proclaimed “Industrial-Strength Natural Language Processing” tool. This post will shed some light on its capabilities outside of named entity recognition.

A note on compression based classifiers

nlp

python

compression

text-classification

This blog post dissects the popular paper titled “Low-Resource” Text Classification: A Parameter-Free Classification Method with Compressors.

Modeling mindsets

statistics

modeling mindsets

python

Illustrations and experiments supplimentary to the excellent modeling mindsets by Christoph Molnar.

Who wrote it first, or: why originality matters?

news

originality

A perspective on the news industry and how to create more relevant news feeds for your customers.

Python dataclasses, a package full of surprises

python

dataclasses

When it comes to serialisation, premature optimisation is worth the effort. A cautionary tale of Python’s dataclass library.

yGuard is now open source - obfuscation easy as pie

java

oss

obfuscation

yGuard, a Java bytecode obfuscation and shrinking tool, has been released openly, thanks to a joint effort with yWorks.