THE PART NOBODY PUTS ON SLIDES

DataScienceGO 2018, San Diego

12–14 October 2018 · San Diego, California · LAB

Three days at DataScienceGO, closing with a main-stage session on the part of deep learning nobody puts on slides: tuning.

DataScienceGO ran across three days in San Diego, from 12 to 14 October 2018. Tarry Singh spoke on the main stage on the Sunday afternoon, the closing day. The venue was a hotel ballroom with two large screens, blue wash lighting, a projected gobo in the lobby and a countdown clock on the confidence monitor facing the stage. The photographs from the session catch it running down through the last eight minutes.

The talk was about tuning. One slide, carrying the dk mark, read "Current Pain With Learning Rates", with the line underneath: "The traditional way of tuning: Try, Try and Try again". Another worked through fuzzy logic using the standard tipping example, if quality is good then tip is high, with the membership functions drawn out on screen. The room was full. Attendees stood along the back wall behind the last row of chairs.

The rest of the weekend followed the same shape. A session on the Saturday covered IBM's Center for Open Source Data and AI Technologies and its work on the enterprise AI lifecycle in open source. On the Sunday at midday a five-person panel sat on the main stage taking questions from a moderator. On the expo floor, event crew in shirts reading WE IGNITE TRANSFORMATION worked the registration and merchandise tables.

The most useful record from the three days is not a slide. It is the group shot taken outside the hotel entrance on the evening of the 14th: speakers and attendees still wearing their conference badges, arms round each other, no stage and no branding anywhere in frame. Conferences are judged on talks. What they actually produce is that.

Every banner in the photographs says only DataScienceGO Conference 2018, so nothing further is claimed about the organiser. The title slide of the session was never photographed, so the talk is described by its content rather than named. The panel and the IBM session belonged to other speakers and are here as context only.

FROM THE FIELD · 08 FRAMES

Wide view of the DataScienceGO main stage, Tarry Singh beside a large screen showing a deepkapha slide titled Current Pain With Learning Rates, subtitled The traditional way of tuning: Try, Try and Try again.

The DataScienceGO ballroom from the back of the room during the session, rows of seated attendees facing two screens that both carry a fuzzy logic slide reading If Quality is Good, then Tip is High.

A five-person panel seated at a draped table on the DataScienceGO main stage, a moderator with a microphone on a stool beside a conference banner, audience and camera operator in the foreground.

Group photograph of nine speakers and attendees outside the conference hotel entrance in San Diego, still wearing DataScienceGO lanyards and badges.

A DataScienceGO session in progress seen from the back of a darkened ballroom, the screen carrying slides on IBM's Center for Open Source Data and AI Technologies.

The DataScienceGO expo and registration floor beside a glass atrium, event crew in navy shirts reading WE IGNITE TRANSFORMATION working the tables.

The DataScienceGO logo projected as a gobo onto the timber wall of the conference hotel lobby, delegates seated below.

MOVING FRAMES · 01

Thirty seconds of the main-stage session, the Current Pain With Learning Rates slide up on the screen and the confidence monitor counting down at the front of the stage.

WITH

DataScienceGO

← EARLIERClimbing the deep learning hill in Chicago LATER →ARiA and LAYGo at NAVER Labs Europe