2025-08

2025-08-29

Pico NeoPixel Driver

In-Memory Computation (IBM Europ)

In-memory Hyperdimensional Computing. Geethan Karunaratne . July 27th 2020

In-Memory Computation (Analog Computation)
Limitation on Scale.
Analog has its limitation in resoultion.
So, Use Digital In between Analog Module .
Brain Solve that by Spiking.
IMC is like a Complex Neuron.

2025-08-28

Event Camera for Pi genx320

genx320

The Redwood Center for Theoretical Neuroscience

History of Redwood Center

The book On Intelligence. (Hawkins in collaboration with Sandra Blakeslee, NY Times)
Development of the Hierarchical Temporal Memory (HTM) model of cerebral cortex, which formed the basis of startup Numenta. (George/Hawkins)
Discovery that spike timing in LGN neurons is phase-locked to retinal oscillations and transmits a parallel stream of information to cortex. (Koepsell/Sommer in collaboration with Judith Hirsch lab, USC)
Development of a novel sparse coding algorithm that accounts for the diversity of receptive fields found in cortical area V1. (Sommer in collaboration with Martin Rehn, KTH Stockholm)
A critical analysis of what is actually understood about the function of cortical area V1. (Olshausen in collaboration with David Field, Cornell)
Development of an information-maximization theory of spike timing dependent plasticity (STDP). (Bell, in collaboration with Lucas Parra, CUNY)
A rigorous comparison of ICA and PCA in terms of actual coding efficiency for natural images. (Bethge)
Analysis of the conditions under which dendritic inputs combine super-linearly within dendrites. (Lauritzen)
Development of a theory of predictive coding in the thalamo-cortical loop. (Softky)
Development of population coding models based on random distributed representation in high-dimensional spaces. (Kanerva)
RNI was funded by the Strauss-Hawkins Trust. Its assets were gifted to UC Berkeley to establish the Redwood Center for Theoretical Neuroscience, and most of its staff transferred there in July 2005. The gift funds an endowment that supports ongoing programs of the Redwood Center, in addition to the graduate program and administration of the Helen Wills Neuroscience Institute.

Numenta

Sparse Distributed Memory

This is an implementation of SDM created by Pentti Kanerva in 1988. SDM is a mathematical model that has some cognitive properties.

It was used to make simulations in my master thesis oriented by Ph.D. Alexandre Linhares. I would like to thanks him and M.Sc. Daniel Chada for their friendship and all their contributions to my thesis.

SDM_AI

TypePad will close at 2025/09/30

So I move my book reading catalog to here.

My Reading

偶爾會看到早期自己的文章會覺得那個人是誰啊？所以保不保存其實那個過去都已經陌生了

2025-08-27

GPT family models parameters sizes

Table 2: Predicted sizes (billions of parameters) of GPT family models.

Model	Parameters
GPT-5 (high)	635
GPT-5 (medium)	330
GPT-5 mini (high)	149
GPT-5 (low)	125
GPT-5 mini (medium)	85
gpt-oss-120B (high)	82
GPT-5 (minimal)	57
GPT-4.1	31
gpt-oss-20B (high)	29
GPT-5 mini (minimal)	27
GPT-4.1 mini	27
GPT-4o (Nov ’24)	26
GPT-5 nano (high)	18
GPT-5 nano (medium)	15
GPT-5 nano (minimal)	8
GPT-4.1 nano	7

Digital Simulation Tool

Digital

Digital is an easy-to-use digital logic designer and circuit simulator designed for educational purposes.

bagel-ai language model

bagel-ai

Neuro-vector-symbolic architectures

Micro-LM

MicroVSA

LeHDC: Learning-Based Hyperdimensional Computing Classifier

2025-08-25

Review

Harp

SyncV Specification

2025-08-22

HDC Use on Control

Intelligent Systems and Control Laboratory (ISCL)

2025-08-20

Applied Brain Research

2025-08-19

Spiking Neuron Models Single Neurons, Populations, Plasticity

Spiking Neuron Models

True Competitor

SpikeGadgets

PDF Crop Margins

MinerU PDF to MD format test

MinerU Github

Example

Generailized Event Camera

Math equation failed conversion.

2025-08-18

Embedded GO

Tiny GO

PPS Technical Considerations

2025-08-15

NeuroChip3 from Fetz Lab

NeuroChip3 Lab

PicoCalc Micropython

Awsome PicoCalc

2025-08-14

Pico use Rust

Pico Pico - Embedded Programming with Rust

PipelineC HDL

PipelineC

A C-like(1) hardware description language (HDL)(2) adding high level synthesis(HLS)-like automatic pipelining(3) as a language construct/compiler feature.

PipelineC YT Channel