TDWG2026
Home / Talks / SAL B
talk · Tuesday 22 September · 16:30–⁠16:35 · SAL B

Robots Rapidly Writing Records: Machine Annotations at Scale with DiSSCo

Soulaine Theocharides · Bots, Bits, and Biodiversity

Recording time 7:49:08–7:53:52Open on Vimeo ↗

Summaries and key moments are machine-produced. Check the recording for the speaker’s exact words.

The short versionDiSSCo can now run adapted machine annotation services over whole datasets, pointing to an image-to-verified-data digitisation pipeline.

Overview

What this was about

Soulaine Theocharides explained that DiSSCo harmonises collections data from many institutions so machine annotation services, adapted once, can serve all partners, avoiding duplicated development. Pilot services include plant organ detection (LeafMachine), links to GBIF/ENA/GeoCASe, georeferencing with GeoPick, and label transcription to Darwin Core using a vision model; DiSSCo can now run services over entire datasets at once. The envisaged pipeline ingests only images, runs services, lets remote experts annotate, and returns verified data to institutions. In Q&A she explained that data come from member countries under CC0 or CC BY.

Why it matters. Shows how shared infrastructure lets institutions reuse AI services at scale instead of building their own.

Key ideas

In the room

  • Digitisation is getting harder to scale; institutions work independently and duplicate effort.
  • DiSSCo brings institutions' data into one structure so services can be reused by all partners.
  • Pilot machine annotation services on the sandbox: LeafMachine plant organ detection, linking to GBIF, ENA and GeoCASe, GeoPick georeferencing, and label transcription into Darwin Core; none developed by DiSSCo, all adapted.
  • New: running services over entire datasets at once rather than per specimen.
  • Pipeline goal: ingest images, run services, remote human experts annotate, verified data returned to institutions (export exists; return in development).
  • Virtual hackathon next month for connecting services; DiSSCo is becoming an ERIC to enable service level agreements.
Jump in

Notable moments

In their words

Transcript

Automatically generated captions can contain mistakes, especially in names and technical terms. Times are relative to the room recording.

Read the transcript ↓
Loading transcript…
Keep exploring

Related talks.

See the map ↗
talk · 22 Sept · 11:45–⁠12:00 · SAL B

Closing the Loop: The DiSSCo Annotation Validation Framework for Trustworthy Data Round-Tripping

DiSSCo proposes impact-based, layered validation so trusted annotations are promoted automatically and only high-impact ones need expert review.

Wouter Addink
talk · 24 Sept · 11:25–⁠11:40 · SAL A

Accelerating specimen identification through Virtual Collections while promoting collaboration

Virtual reference collections with DOIs let experts identify and curate dispersed digital specimens remotely instead of shipping material or travelling.

Melanie de Leeuw
talk · 22 Sept · 11:27–⁠11:39 · SAL A

HerbAudit: Validating AI-Driven Herbarium Transcriptions

HerbAudit shows AI herbarium transcription reaching ~94% accuracy and provides a fair, field-aware way to benchmark it.

Dilara Ağacık
Programme abstract

Robots Rapidly Writing Records: Machine Annotations at Scale with DiSSCo

Soulaine Theocharides

In the face of the biodiversity crisis, digitization efforts in natural science collections need to rapidly increase. Scaling up these efforts requires the support of the community - and machines. DiSSCo, the Distributed System of Scientific Collections, enables this through annotation and harmonization of the data.

When natural science collection datasets are ingested into DiSSCo, specimens are transformed into Digital Specimens, assigned a Persistent Identifier, harmonized to a shared data model, and become and become fully actionable by machines as discrete knowledge units.

Through this process, DiSSCo enables community curation through annotations by humans and machines. With Digital Specimens harmonized to one data model across host institutions, DiSSCo is uniquely situated to host “plug and play” machine services to act on any specimen in the DiSSCo infrastructure. Instead of being confined to a single dataset they were designed for, Machine Annotation Services (MAS) adapted for DiSSCo can act on any Digital Specimen, breaking institutional silos and furthering collaboration.

For the first time, DiSSCo MASs have been tested at scale, running on all digital objects in a dataset as they are ingested into the DiSSCo infrastructure. These pilot services link all Digital Specimens in a given dataset to counterparts that can be found in GBIF, ENA, or BOLD. With the pilot successful, there is now an opportunity to run more complex services (such as label transcription using large language models) at scale on natural science collection data in DiSSCo.

This lightning talk gives an overview of the MAS-on-Ingestion feature in DiSSCo and what that means for transforming collection data.

Submitted to the conference programme before the talk, so the talk itself may differ.