Lmst

Origin 006 Core đạt bước tiến lớn: xử lý 100.000 điểm dữ liệu trong 14.73s tại 200D – không dùng GPU, không backprop. Tốc độ 6,788 điểm/giây, độ trễ trung bình 147μs, nén dữ liệu 50.04% bằng hình học định hướng xác định. Chạy trên CPU Colab thông thường. Purity Mode giúp duy trì cấu trúc trong không gian cao chiều. Mở ra hướng mới cho xử lý dữ liệu hiệu năng cao, tiết kiệm năng lượng. #AI #MachineLearning #Origin006 #DeterministicAI #HighDimensional #LLM #TríTuệNhânTạo #HọcMáy #XửLýDữLiệu

https

I like that someone did a followup to 3blue1brown video on near-orthogonality:

Beyond Orthogonality: How Language Models Pack Billions of Concepts into 12,000 Dimensions

https://nickyoder.com/johnson-lindenstrauss/

> This research suggests that current embedding dimensions (1,000-20,000) provide more than adequate capacity for representing human knowledge and reasoning. The challenge lies not in the capacity of these spaces but in learning the optimal arrangement of concepts within them.

#highdimensional #math

🧠 New comprehensive review on #LowDimensional #embeddings of #HighDimensional data. Discusses how #dimensionalityreduction helps visualizing, exploring, and #modeling #ComplexSystems. From #PCA to #tSNE, #UMAP & #NeuralNetworks: Excellent overview paper👌

🌍 https://arxiv.org/abs/2508.15929

#CompNeuro #MachineLearning #DataVisualization

Figure 4: 2D embeddings of 23 800 cells from the mouse cortex (Tasic et al., 2018). Colors correspond to transcriptomic cell types,
taken from the original publication. The first two principal components explained 49.1% of the variance of the preprocessed data.
As Laplacian eigenmaps had many almost-overlapping points, they are shown with larger semi-transparent markers.

Figure 6: 2D embeddings of 3 450 human genotypes from 26 global populations (The 1000 Genomes Project Consortium, 2015).
Colors represent the sampling population. The first two principal components together explained 5.8% of the total variance. For
population abbreviations used to annotate the UMAP embedding, see the original publication. As Laplacian eigenmaps had many
almost-overlapping points, they are shown with large semi-transparent markers.

Let's start designing a new course for applied mathematics students in #UCLouvain, #EPL on high dimensional data analysis with 3 wonderful reference books #inverseproblem #highDimensional #statistics #optimization #Sparsity #teaching

Picture of the 3 books I'm going to use for this course:

Vershynin, R. (2018). High-dimensional probability: An introduction with applications in data science (Vol. 47). Cambridge university press.

Wright, J., & Ma, Y. (2022). High-dimensional data analysis with low-dimensional models: Principles, computation, and applications. Cambridge University Press.

Wright, S. J., & Recht, B. (2022). Optimization for data analysis. Cambridge University Press.

Single cell omics #singlecellomics from bench to bedside: conference and workshop at KU Leuven in Belgium. Two days of technology and application in October give a deep insight in the nuts-and-bolts of the current state of the art. Meet Abhishek Garg, Yvan Saeyes, Sabine Tejpar, Baki Topal, Jessica Roelands, Steven de Vleeschouwer and more @cyrilpedia @chfloudas #biology #transcriptomics #proteomics #multiplexing #tissue #clinics #standardization #data #highDimensional
https://gbiomed.kuleuven.be/english/research/50000622/single-cell-course

Looks like advances in neural computing theory might have analog computing back with a vengueance. 🤯

#reservoircomputing #analog #computing #highdimensional #computingspace

https://www.nature.com/articles/srep22381

@TedUnderwood "intuition about #highDimensional space" *grin*

Does anyone know of a research project which explored structuring oppositions in #highdimensional spaces constructed from fictional narratives? #digitalhumanities

I have come across this only from an approach in art history, where the artworks represented in the vector space were distributed in clusters, and therefore represented oppositions such as naturalism vs. impressionism.

See
https://digital-narratives.versae.es/#

Can someone with better intuition about #highDimensional space tell me if this is right?

#highDimensional

Client Info