EurIPS 2025

Tue, 02 Dec 2025 11:33:55 +0100

I attended EurIPS 2025 in Copenhagen, the first edition of the European version of NeurIPS. My notes are mainly about topics related to my current work, that is, time series analysis and biosignals.

EuroRust 2025

Sat, 11 Oct 2025 11:58:15 +0200

I attended my first Rust conference, EuroRust 2025 in Paris.

clippy settings recommendations, including pedantic or undocumented_unsafe_blocks (structured safety documentation) for unsafe code
best practices for Rust development by jhpratt (slides, also see his related talk on compiler-driven development)
inner workings of the Rust compiler, compiler settings and their trade-offs including inlining, monomorphization link-time optimizations (LTO) by noratrieb (slides, Rust performance book)
best practices for error handling: for anticipated errors, prefer thiserror for libraries to expose concrete and structured errors, and Result type and anyhow for application code, for unanticipated errors use panic! (programmer mistake)
criterion and flamegraphs for profiling
inner working of atomic types for concurrency and memory ordering
prefer concrete types (e.g. enums) over generics and traits, for speed due to monomorphization and readability, especially in application code
presenterm for making presentations in your terminal
embassy for embedded systems
bevy game engine
the concept of run-time reflection and and Rust’s compile-time solutions based on procedural macros like serde’s derive, e.g. when you add #[derive(Serialize, Deserialize)] to a struct, you are using a procedural macro that runs when you compile your code; the macro inspects the definition of your struct (e.g., the field names and types), and based on this inspection automatically generates the implementation of the core Serialize and Deserialize traits for your specific type
triemap, a tree-based data structures primarily used for efficient storage and retrieval of associative data where the keys are strings or sequences, e.g. for fast prefix-based lookup
foreign function interfaces (FFI) in Rust for using functions from other programming languages, particularly C (e.g. using extern)
Valgrind for profiling and memory debugging
data engineering
- OLAP (analytical)
  - batch operations involving complex queries, aggregations and scans across many rows but few columns
  - primarily columnar data format, e.g. Parquet/Arrow
  - data warehouse solutions like Delta Lake and Snowflake
- OLTP (operational)
  - frequent inserts, updates, deletes, and fetching full records (e.g. banking transactions, user profiles)
  - row-based format (e.g. PostgreSQL)
  - document-based/NoSQL databases (e.g. MongoDB), better for complex entities to avoid complicated SQL joins

Videos will be available at: https://www.youtube.com/@eurorust.

Pydata Paris 2025

Thu, 02 Oct 2025 11:58:15 +0200

quarto for publication-ready reports and presentations (see e.g. https://github.com/paddyroddy/talks)
an overview of the array API standard by Lucas Colley
optimal transport theory and Python library (https://github.com/PythonOT/POT)
free-threaded Python rollout (see https://labs.quansight.org/blog/free-threaded-python-rollout, https://lwn.net/Articles/1025893/)
conflict-free replicated data type (CRDT) for lock-free concurrent read/write access, they don’t need locks because they accept conflicts but define rules (“merge” operations) that resolve conflicts (e.g. used collaborative text editors like Google Docs) (see https://read.thecoder.cafe/p/crdt)
met the maintainer of tslearn, @charavelg
Introductory talk on prediction intervals and quantile regression based on scikit-learn, but interface seems incomplete, more complete interfaces exists

Talks are available on here.

Conferences on Markus Löning

EurIPS 2025

EuroRust 2025

Pydata Paris 2025