Skip to main content

πŸ“… Meetings

Status:Preliminary
Tags:MeetingProject Log
Last modified:

Meeting registry

This page is the project meeting calendar. Meetings are stored in site/src/data/meetings.js and rendered here. Each entry can record a date, status, meeting type, attendees, links, a short markdown notes preview, and a pointer to a full markdown note page. Attendees should use registered people IDs so the chips link back to the project roster.

May 2026
SunMonTueWedThuFriSat
26
27
28
29
30
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
21
22
23
24
25
26
27
28
29
30
31
1
2
3
4
5
6

Meetings in May 2026

MTG-2026-05-20-STATUSCompleted

Alan, Seda, Manu sync and next steps

2026-05-20working-session

Presented documentation. Discussed batch norm and global-throttling planning.

@Manu B. Valentin@Alan Guo@Seda Ogrenci
  • Andrew/Darell tested batchnorm. It improves accuracy in software but still suffers from drift (good for ENABOL).
    • They will now freeze dense and try to train conv layers only.
  • Alan found that kappa throttling is uneven across layers, this distorts learning.
    • He will work on the backpass for BN (even if not training it, we need propagated gradients).
    • He also needs to consolidate the logging to always keep the pre-drift baseline, post-drift initial + recovered (3) points.
  • We basically need 3 tests in the mid-term (without dense training):
    • Only training conv layers (no batchnorm training).
    • Only training batchnorm layers (no conv training).
    • Training both conv and batchnorm layers together.
  • Manu presented the global throttling. Seems effective.
    • Next steps are to do all the rest of ablation tests with throttling.
    • Pull hls4ml and run first csim with synthesizable controller for global throttling.
Current StatusTasks

Registry Format​

A meeting entry is intentionally small:

Meeting record shape
  1. meeting:
  2. id: MTG-YYYY-MM-DD-SLUG
  3. title: short readable title
  4. date: YYYY-MM-DD
  5. status: planned | completed | canceled
  6. type: working-session | research-planning | project-update | handoff
  7. attendees: [mbvalentin, alan-guo]
  8. summary: one short paragraph
  9. notes: markdown preview shown in the meeting card
  10. notesDoc:
  11. label: 'Click here for full meeting notes'
  12. href: /docs/meetings/YYYY-MM-DD-meeting-slug
  13. links:
  14. docs:
  15. - label: Related page
  16. href: /docs/...
Source of truth

The calendar registry stays as the structured index. Longer notes should live as dedicated markdown pages under site/docs/meetings/, with the registry entry linking to them through notesDoc.