Conference record: 3rd Descriptive Science on Data (day 1)

5-dimension data visualization

  • Big data in Materials science, 5D (3D + t + E) time, energy
  • Topology of Mine tracking history (continous data)

3D rotation + 4D rotation

  • Raw data(4, 1) -> (3, 4)R(1, 4) = (3, 1)

data=(xyzw)−>(1,0,0,10,1,0,10,0,1,1)R(x,y,z,w)=(x~y~z~)\mathrm{data} = \begin{pmatrix} \mathrm{x} \\ \mathrm{y} \\ \mathrm{z} \\ \mathrm{w} \\\end{pmatrix} -> \begin{pmatrix} 1, 0 , 0, 1 \\ 0, 1, 0, 1 \\ 0, 0, 1,1 \\\end{pmatrix} R \begin{pmatrix} \mathrm{x,y,z,w} \end{pmatrix} = \begin{pmatrix} \tilde{x} \\ \tilde{y} \\ \tilde{z} \\\end{pmatrix}

  • Visualization on energy fading
    (X,Y)/rightarrow(X, Y) /rightarrow

  • Function

    • VR device: data panel (with reset)
    • VR device: params setting UI

Topological node2vec (github access)

paper with code: https://paperswithcode.com/paper/topological-node2vec-enhanced-graph-embedding
github: https://github.com/killianfmeehan/topological_node2vec

  • Task of finding neighborhoods

  • example of how it works
    Nl=l,r=r(A)=A,A,A,A=(0,0,1/3,1/3)N_{l=l, r=r}(A) = {A, A, A, A} = (0, 0, 1/3, 1/3)

    • l = length of each walk
    • r = # of walks

  • Motivation: RQ = how to connect clips of DNA

    • original data structur (Hi-C data): w-x x-y y-z z-w
    • want to infer w-x-y-x-z-w
    • BUT it is extremely difficult to do so
      -> introduce topology

  • How to reach

    • original N2V
    • totally new function with topological

  • Future work

    • how to distinguish?

High dimension data analysis

  • How can quantify CF alignments in CFRP
  • XAFS imaging on Ion-battery material
    In this work, visualization shows that when discharge, different energy status of each point
    • Fe(II) -> Fe(III) charge and discharge process different with spatial
    • Explanation: Ion motion theory? -> reaction distribution
    • On NiO/Ni battery material: same pattern with traditional Fe/Li based battery
      -> visualization and Explanation on inhomogeneity of reaction distribution
  • Forging on Fe materials -> how to visualize XAFS
    • New tech on enlarging the view of filed of micro structure of Fe
      • Valance = color ; Scale = um
    • Classification on new measurement (measurement 10hr; analysis 20hr)
      • Classification with Pearson Similarity
      • Task: Segmentation on SEM-EDX

Interval Approximation

arXiv: https://arxiv.org/abs/2308.14979
Intro: detection on data’s noise (multiple parameterization)
RQ:

morphism f
A minimal interval resolution of M is sequence

  • f restricted to each indecomposable summand of X is an injection

Infinite MDS

ref arXiv: 2201.09385v5
RQ: if X is not well symmetric?

  • infinite MDS on HjH_j -> well defined
  • in multiple variables: well defined?

Machine and human language

See Slide provided by Prof. Fukui: Slide

RQ: The shape of the language ‘vector?’

  • One single word existence effect on contexts-> Transformer?
    Theme: NLP and LLM
  • Introduction of previous studies (How NLP evolve)
    • On application:
      • Support human (interpersonal) communication
      • Text completion, bot chats (human-computer interaction)
    • On implication
      • Generalization of human and computer language
  • How to build a chatbot based on LLM
    • Prediction of the next word
    • Loss function: cross-entropy
    • LLM for NLPer
      • Terminal of NLP?
      • LLMs also get limited
        • 1st language learning
        • Memorization and Generalization
  • How far? and what kind of mechanism it reached?
    • Emergent ability theory -> will LLM reach intelligence?
    • Abstraction ability? in many dimensions
  • Machine and human interaction
    • LLM is language but not thoughts

Keywords: #conference #records #math