Large decision models and combinatorial search
Notes on using Jev and other decision models as branching oracles: Sudoku, A* tie-breaking, and the experiments needed to tell useful guidance from expensive guesswork.
Writing / Ideas in progress
Research notes
Research and essays on AI, physics, mathematics, finance, neuroscience, and philosophy, alongside practical computing notes.
135 articles
What a research PR to skfolio 1.0.0 actually contains
A maintainer-facing cut: rebase universal_portfolio onto v1.0.0, the in-list and out-list, tests, gallery, naming, and the follow-up issues that must not ride along.
First-order OCO as a portfolio estimator
How the universal_portfolio branch implements COMID, EG/OGD, follow-the-loser PA, BCRP as a MeanRisk comparator, and which pieces belong in a first skfolio 1.0 PR.
Incremental moments, full solves
skfolio 1.0 online_predict and MeanRisk.partial_fit carry exponentially weighted moments and still rebuild the convex program. That is streaming Markowitz, not online convex opt...
Binding cross-validation to solver structure
How skfolio-accelerate compiles a CV plan once, slides empirical Grams, and reuses OSQP/Clarabel topology without changing the MeanRisk investment problem.
MeanRisk as a parametric convex program
What skfolio 1.0 MeanRisk actually solves, why every fit rebuilds a CVXPY graph, and why WalkForward is a trajectory through θ rather than T unrelated estimators.
From solving convex programs to tracking them
A four-regime map from batch MeanRisk through skfolio-accelerate and 1.0 streaming MVO to true OLPS, written as study notes for a skfolio 1.0 PR.
Memory in the age of AI agents: forms, functions, and dynamics
A readable tour of the Hu et al. survey on agent memory—formalization, the forms/functions/dynamics taxonomy, and the methods that matter most.
Un agente non è un modello: self-agency, ambiente e strange loop negli LLM
La coscienza artificiale non si decide dai discorsi in prima persona. Una proposta di ricerca: studiare la self-agency nel circuito ricorsivo fra agenti, ambiente, memoria e mod...
Portafoglio gestito con tassazione annuale o buy and hold? Simulazione Montecarlo
Confronto matematico e Monte Carlo tra la Gestione Patrimoniale Moneyfarm (tassazione annuale sul risultato di gestione) e un conto trading buy-and-hold che replica lo stesso pr...
Feynman-Kac theory of reasoning models
A research programme for measuring commitment transitions, gauge-invariant observables, and universality in the free-energy landscape of autoregressive generation.
SAT solver tramite una rete neurale
Introduzione ad una variante del modello NeuroSAT usato nel progetto RYS: letterali, clausole, matrici di incidenza, stati latenti, output e rumore stocastico
La mente che conta fino a dieci
Permutation City, i transformer e la fenomenologia del tempo: perché la cognizione costa poco, ma un sé che vive nel tempo resta legato a un corpo.
A controlled testbed for Repeat-Your-Self — SAT solving, message passing, and a predictive rho/phi theory
Experiments involving the extension of the RYS approach to simpler transformers and neural networks with a supporting correlation based theory
Deep Equilibrium Models as a model for RYS
Using deep-equilibrium-models can be a way to improve the theory behind the CKA-based prediction about which are the best sequences of layers to repeat
Fractal Connectivity Scaffold Networks
A running note to keep track of experiments about fractal connectivity networks hypothesis
Hippocampal scaffold networks
We test two brain-inspired architectures — Hippocampal Scaffold Network (HSN) and Cerebellar Expansion Network (CEN) — that trade a tiny number of learned parameters for a large...
Multiscale structured recurrence with fractal connectivity networks
Can we exploit multiscale fractal connectivty structures and echo-state networks to devise a new low-parameters alternative to the MLP in large neural networks?
Similarity of neural networks representations
We develop a general theory for describing the connectivity among layers in neural networks inspired from the RYS (repeat-your-self) approach to improve neural networks increasi...
How skip connections define graphs in deep networks
How to extend the RYS (repeat-your-self) approach of repeating LLM layers with a centered-kernel-alignment metric and predict structural blocks to repeat
Functional synthetic LLM connectome analysis
How we can link transformers way of processing information to biological brains
PLP Inference Audited Against Murphy
How the current implementation of PLP covers the Kevin Murphy's book
The Master Equation of a Messy World
hysics progressed by compressing the universe into elegant, predictive equations, anticipating phenomena long before they could be measured. But domains like human language and ...
Kolmogorov-Arnold theorem, KANs, and the `eml` operator
A single exp-minus-log gate looks like a frozen Kolmogorov-Arnold layer. KANs learn the one-dimensional functions, while EML composes one fixed binary primitive.
Amortized structural variational inference for probabilistic language programming
How replacing linear agent loops with chain graphs over NAND gates leads to a principled ELBO formulation of amortized structured variational inference and why that matters for ...
The continuous Sheffer stroke is all you need
How we can use the EML operator in machine learning to study probabilistic logical learning
From brain to disk: how I built a voice-activated second brain for $0
Some ideas to build a voice activate second brain as an LLM-wiki in Andrej Karpathy style
Interpretations for the Kullback-Leibler divergence, or relative entropy
Here we provide multiple intuitions about the meaning of the Kullback-Leibler divergence in many fields
Latent traces and Landau free energy
Chain-of-thought as latent-variable inference, with a conjectural grand-canonical extension.
Chain-of-thought as grand-canonical inference in energy-based language models
Jaynes, Blondel's soft Bellman recursion, and a grand-canonical view of latent reasoning traces.
Statistical mechanics of decision making
From Jaynes and the soft Bellman equation to a grand-canonical reading of chain-of-thought, with fugacity as reasoning effort.
From hard to soft operators: between machine learning and statistical physics
Computer scientists did not study statistical mechanics enough to recognize the strong similarities betweeen hard (combinatorics, microcanonical) and soft (slack-constraints, ca...
Random rooted trees, continuation free energy, and the Diligent Learner
A unified theory of delayed commitment, decomposition tax, and tool building in scaffolded inference.
Short-prefix soft values: a null result
A small experiment shows that easy tasks saturate too quickly to test the predictive power of short-prefix soft values.
Recursive decomposition with a continuation policy
Adding an abstain action to a recursive scaffold shows how delayed commitment can help, but only in a narrow ambiguity regime.
MellowMax, Doob's h-transform, and the intensive geometry of tool-use agents
A theoretical framework for optimal tool selection in Large Reasoning Models using intensive soft values and dynamic programming.
Soft values, symmetry breaking, and random rooted trees
A conjectural branching law for scaffolds from Blondel's soft values and Kappen's path-integral control.
Inference-Time Steering as a Discrete Schrödinger Bridge
Formalizing LLM self-correction, Doob's h-transform, and Wasserstein gradient flow.
PLP is inference-time approximation of free energy
PLP as inference-time free energy / log-partition over traces.
Test-time inference, language models and energy based models
PLP, energy-based models, and test-time inference over LLM traces.
Running Claude Code via OpenRouter in Docker
Claude Code in Docker + OpenRouter; Debian-slim, keys at runtime.
Unleash Claude Code CLI with OpenRouter: Free AI Coding Power!
Claude Code with OpenRouter and free models; env and auth notes.
Scaffolding is all you need
When recursive LLM scaffolds improve reliability—and when they cannot.
Probabilistic Language Programming, turning craft into solid science
PLP: semantics for LLM scaffolds—traces, proposals, verifiers, branching.
Intelligenza Artificial Neurosimbolica
Come integrare sistemi simbolici e non simbolici in una AI ibrida può risolvere i problemi di entrambi i mondi
Introducing skfolio new online convex optimization module
skfolio online convex optimization and sequential portfolio allocation.
Hard and soft analogies in machine learning
Jumping from hard to soft constraints defines a jump from pure combinatorial to boltzmann statistics
Configuration-model Surprise for community detection
Surprise vs configuration null for community detection; stubs and pairings.
Running LM-Studio headlessly on Ubuntu 24.04 with CUDA
Headless LM Studio on Ubuntu 24 + CUDA + systemd.
Unlocking the Power of DeepSeek R1 via HuggingFace's TogetherAI
DeepSeek R1 from the CLI via Together AI and Hugging Face.
Prompt Chaining with Foundational models
Prompt chaining: multi-step LLM workflows and handoff patterns.
Creating the How2 Function for Natural Language Bash Commands with LLM
Bash how2: natural language to shell commands via llm CLI.
Multilabel classification with Parabel and efficient at-k metrics
Parabel multilabel classification, scikit-learn API, Precision@k / Recall@k.
PySpark initialization outside the Pyspark shell
PySpark with findspark: SparkSession without the pyspark shell.
Installing XFormers on Mac M1/M2
Build XFormers on Apple Silicon: Homebrew libomp/llvm and clang flags.
Augmentation of documents for neural networks training
Augmentation of documents for neural networks training.
Slicing rolling expanding windows over multiple pandas objects
Slicing rolling expanding windows over multiple pandas objects.
Dependencies needed for detectron2 on mac os
Detectron2 can be installed on MacOs on M1 apple silicon
How to enable colab for longer runtimes by keeping it active
How to enable colab for longer runtimes by keeping it active.
How to download google sheet data to Pandas
How to download google sheet data to Pandas.
Better graphics with matplotlib and seaborn
Matplotlib and seaborn defaults for publication figures: SVG, Helvetica, tight layout.
Changepoint detection on huge grouped dataframes with ruptures and PySpark
Changepoint detection on huge grouped dataframes with ruptures and…
How to make HDBScan an inductive clustering method
How to make HDBScan an inductive clustering method.
How to convert networkx graphs to graph-tool
How to convert networkx graphs to graph-tool.
Finding and deleting memory hungry temporary variables in jupyter notebooks
Finding and deleting memory hungry temporary variables in jupyter…
Versioning data files for machine learning projects with DVC
Versioning data files for machine learning projects with DVC.
Prefixspan algorithm for frequent subsequences visualization
Prefixspan algorithm for frequent subsequences visualization.
How to merge Dataframes in spark as in Pandas
How to merge Dataframes in spark as in Pandas.
Installing Jupyter with a Scala + Spark kernel
Installing Jupyter with a Scala + Spark kernel.
How to remove all logging information in scala spark
How to remove all logging information in scala spark.
Eigenvectors from eigenvalues a numpy implementation
Eigenvectors from eigenvalues a numpy implementation.
Sampling uniform spanning forests with Wilson algorithm in Python
Sampling uniform spanning forests with Wilson algorithm in Python.
The link between machine learning and statistical physics
The link between machine learning and statistical physics.
Quantum statistical mechanics of complex networks in networkqit
Quantum statistical mechanics of complex networks in networkqit.
Exercises for quantitative interviews
prepare for a quant interview, if it still matters in 2026!
Visual inspection of python profiler output with gprof2dot
Visual inspection of python profiler output with gprof2dot.
Showing brain parcellation in Python with nilearn and some hacking
Nilearn surface plots for brain parcellation and ROI visualization.
Faceting images or generic plots with Seaborn and Python matplotlib
Faceting images or generic plots with Seaborn and Python matplotlib.
Install latest igraph 0.7.1 for Python3 on Ubuntu
Install latest igraph 0.7.1 for Python3 on Ubuntu.
Weighted graphs from adjacency matrix in graph-tool
Weighted graphs from adjacency matrix in graph-tool.
Random matrix ensemble spectral density and the average resolvent
Random matrix ensemble spectral density and the average resolvent.
Some considerations about the spectral entropies framework
Some considerations about the spectral entropies framework.
Why probabilistic programming matters
Probabilistic programming as seen from the past (a paper appeared before 2014!)
A Full Undergraduate Course in Econometrics
Notes taken from the online course by Ben Lambert about econometrics and statistics
The Enhanced weighted random graph model (EWRG)
The Enhanced weighted random graph model (EWRG).
Plugging the weighted random graph into Surprise
Plugging the weighted random graph into Surprise.
Statistical mechanics of networks - Park and Newman model of hidden variables
Statistical mechanics of networks - Park and Newman model of hidden…
Eigenvalue spectra of modular matrix techniques and calculations
Eigenvalue spectra of modular matrix techniques and calculations.
How to stop with CTRL C a running mex file in Matlab
How to stop with CTRL C a running mex file in Matlab.
How to grant specific directory access to users in linux
How to grant specific directory access to users in linux.
How to create a large swap file in Ubuntu if system runs out of memory.
How to create a large swap file in Ubuntu if system runs out of memory.
TAB is not working with XFCE4 after installation from Ubuntu for use with remote desktop
TAB is not working with XFCE4 after installation from Ubuntu for use…
Matlab and CUDA gpudevice freezes on Ubuntu 14.04 with GTX 1070
Matlab and CUDA gpudevice freezes on Ubuntu 14.04 with GTX 1070.
Computing Euler angles from 3x3 rotation matrix in Matlab
Computing Euler angles from 3x3 rotation matrix in Matlab.
Computing hypergeometric probability efficiently in C++
Computing hypergeometric probability efficiently in C++.
How to start a SSH daemon on Windows without agonizing pain
How to start a SSH daemon on Windows without agonizing pain.
How to upgrade R to the latest version on Ubuntu 14.04
How to upgrade R to the latest version on Ubuntu 14.04.
A code for the absolute orientation problem with Umeyama algorithm in Python
A code for the absolute orientation problem with Umeyama algorithm in…
Delaunay triangulation and beautiful visual effects in Latex
Delaunay triangulation and beautiful visual effects in Latex.
Comparing spectral densities of random graph models.
Comparing spectral densities of random graph models.
Matlab R2016a crashes on Ubuntu 16.04 with NVidia 361 drivers
Matlab R2016a crashes on Ubuntu 16.04 with NVidia 361 drivers.
Fix black screen after upgrade from Ubuntu 14.04 to Ubuntu 16.04
Fix black screen after upgrade from Ubuntu 14.04 to Ubuntu 16.04.
Launching-matlab-in-background-correctly-in-linux
Launching-matlab-in-background-correctly-in-linux.
How to install latest R on servers under proxy
How to install latest R on servers under proxy.
Get the exact print size of a PDF in inches
Get the exact print size of a PDF in inches.
Forwarding-X11-maintaining-connection-with-xpra
Forwarding-X11-maintaining-connection-with-xpra.
No articles match your search and selected topic.