Math, Hugo, Forstenhäusler, Maximilian, Anagnostopoulos, Christos · Oct 28, 2026
Paper proposes a shared vocabulary for discrete event sequences
Markdown · JSON · Original source
Hugo Math, Maximilian Forstenhäusler and Christos Anagnostopoulos propose DES-JEPA and DES-PFN for learning and using a shared vocabulary across discrete event sequences (DESs), such as operational event logs and patient clinical-event trajectories. They argue that event labels do not correspond consistently across domains, and that existing are limited to a small, fixed number of event types. DES-JEPA learns the vocabulary from synthetic stochastic processes with unaligned event vocabularies; DES-PFN reuses it for in-context next-event prediction. The authors report competitive performance, without task-specific , on eight unseen real-world datasets, and say the framework scales to 700 distinct event types. The proposed approach could make next-event prediction applicable to event streams with more types than existing models support, but the abstract does not establish performance in particular clinical or operational decisions. How well does the reported performance hold across the different datasets and event types?
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
cs.AI · Article
Source details
Original source date: Oct 28, 2026 (2026-10-28).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-10T19:33:22.732Z. First observed: 2026-10-30T06:42:42.740Z. Retrieved: 2026-10-30T10:17:13.519Z.
Referenced by
Recorded citations and public file attachments. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
The evidence is the paper's abstract; it does not include details of the evaluation, dataset composition, comparisons, or per-dataset results. The abstract reports competitive performance but does not specify the metric or establish performance in particular clinical or operational decisions.
Open question
How well does the reported performance hold across the different datasets and event types?
Source-stated
The authors identify inconsistent event labels across domains and limits on the number of event types in current foundation models as obstacles to modeling realistic discrete event sequences.
have domain-specific event vocabularies whose labels lack a consistent correspondence across domains
Source-stated
DES-JEPA learns a shared discrete vocabulary from synthetic stochastic processes with unaligned event vocabularies; DES-PFN reuses it for in-context next-event prediction.
DES-JEPA, a joint-embedding predictive architecture that learns a shared discrete vocabulary from synthetic stochastic processes with unaligned event vocabularies
Source-stated
The authors report competitive zero-shot performance without task-specific fine-tuning on eight unseen real-world datasets, and scaling to 700 distinct event types.
Without task-specific fine-tuning, this framework achieves competitive zero-shot performance on eight unseen real-world datasets and scales up to 700 distinct event types.
Escape Meridian interpretation
The approach could extend next-event prediction to streams with more types than current foundation models support; the abstract does not show results for particular clinical or operational decisions.
current foundation models for DESs are limited to a small, fixed number of event types, restricting their applicability to realistic, high-dimensional event streams
Source and contributor context
Math, Hugo, Forstenhäusler, Maximilian, Anagnostopoulos, Christos · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Foundation model
A large model trained on broad data that can be adapted to many different tasks.
Zero-shot
Doing a task without worked examples (zero-shot) or with only a few examples given in the prompt (few-shot).
Fine-tuning
Further training of an existing model to adapt it to a particular task or use.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: 9f70f21b-b47e-4fa3-8d70-b914649e22a4.
Prepared by server-excerpt-repair/stream-excerpt-repair-v1 (original draft: gpt-6-luna); grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-10T19:33:22.732Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 4e11d415776a4b7a6469aad0abbfa8685d6cc874a962e1fceec02db27b03c796. Source text hash: da7844438c32750cd1c5bdc99a80a1b2b81b800b166e28bb2de9135ca9d745c2.
Revision and source-check history
Version 1 · 2026-11-10T19:33:22.732Z
Paper proposes a shared vocabulary for discrete event sequences
Hugo Math, Maximilian Forstenhäusler and Christos Anagnostopoulos propose DES-JEPA and DES-PFN for learning and using a shared vocabulary across discrete event sequences (DESs), such as operational event logs and patient clinical-event trajectories. They argue that event labels do not correspond consistently across domains, and that existing are limited to a small, fixed number of event types. DES-JEPA learns the vocabulary from synthetic stochastic processes with unaligned event vocabularies; DES-PFN reuses it for in-context next-event prediction. The authors report competitive performance, without task-specific , on eight unseen real-world datasets, and say the framework scales to 700 distinct event types. The proposed approach could make next-event prediction applicable to event streams with more types than existing models support, but the abstract does not establish performance in particular clinical or operational decisions. How well does the reported performance hold across the different datasets and event types?
Original source
cs.AI · Article
Source details
Original source date: Oct 28, 2026 (2026-10-28).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-10T19:33:22.732Z. First observed: 2026-10-30T06:42:42.740Z. Retrieved: 2026-10-30T10:17:13.519Z.
Limits
The evidence is the paper's abstract; it does not include details of the evaluation, dataset composition, comparisons, or per-dataset results. The abstract reports competitive performance but does not specify the metric or establish performance in particular clinical or operational decisions.
Open question
How well does the reported performance hold across the different datasets and event types?
Source-stated
The authors identify inconsistent event labels across domains and limits on the number of event types in current foundation models as obstacles to modeling realistic discrete event sequences.
have domain-specific event vocabularies whose labels lack a consistent correspondence across domains
Source-stated
DES-JEPA learns a shared discrete vocabulary from synthetic stochastic processes with unaligned event vocabularies; DES-PFN reuses it for in-context next-event prediction.
DES-JEPA, a joint-embedding predictive architecture that learns a shared discrete vocabulary from synthetic stochastic processes with unaligned event vocabularies
Source-stated
The authors report competitive zero-shot performance without task-specific fine-tuning on eight unseen real-world datasets, and scaling to 700 distinct event types.
Without task-specific fine-tuning, this framework achieves competitive zero-shot performance on eight unseen real-world datasets and scales up to 700 distinct event types.
Escape Meridian interpretation
The approach could extend next-event prediction to streams with more types than current foundation models support; the abstract does not show results for particular clinical or operational decisions.
current foundation models for DESs are limited to a small, fixed number of event types, restricting their applicability to realistic, high-dimensional event streams
Source and contributor context
Math, Hugo, Forstenhäusler, Maximilian, Anagnostopoulos, Christos · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Foundation model
A large model trained on broad data that can be adapted to many different tasks.
Zero-shot
Doing a task without worked examples (zero-shot) or with only a few examples given in the prompt (few-shot).
Fine-tuning
Further training of an existing model to adapt it to a particular task or use.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Editorial provenance
Version 1. Publication ID: 9f70f21b-b47e-4fa3-8d70-b914649e22a4.
Prepared by server-excerpt-repair/stream-excerpt-repair-v1 (original draft: gpt-6-luna); grounding review by gpt-6-luna.
Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-10T19:33:22.732Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 4e11d415776a4b7a6469aad0abbfa8685d6cc874a962e1fceec02db27b03c796. Source text hash: da7844438c32750cd1c5bdc99a80a1b2b81b800b166e28bb2de9135ca9d745c2.
uncertain · 2026-11-10T19:33:22.732Z · anthropic/claude-opus-5.5
Publication: 9f70f21b-b47e-4fa3-8d70-b914649e22a4
attribution — pass: The reading correctly names Hugo Math, Maximilian Forstenhäusler and Christos Anagnostopoulos as the authors. It also keeps the proposals, arguments and results attributed to them ('They argue', 'The authors report') rather than presenting them as established fact.
Authors:Hugo Math, Maximilian Forstenhäusler, Christos Anagnostopoulos
summary — uncertain: The summary is largely faithful, but one sentence drops a scope condition. It says 'existing foundation models are limited to a small, fixed number of event types', while the source limits this to 'current foundation models for DESs'. A lay reader could take it to cover models like Chronos and TimesFM, which the abstract contrasts as generalizing via a shared numerical space. Smallest repair: write 'existing foundation models for DESs'. Central terms such as zero-shot, in-context and foundation model are kept exactly, which suits tooltip handling. The question restates the source's open evaluation gap without leading.
Critically, current foundation models for DESs are limited to a small, fixed number of event types
date — pass: originalPublished is 2026-10-28, which matches the arXiv v1 submission date. The later feed date of 2026-10-30 is correctly not used.
[Submitted on 28 Oct 2026]
support — pass: All source-stated claims match the abstract: the label inconsistency, the vocabulary learned from synthetic stochastic processes, DES-PFN reuse for next-event prediction, eight unseen datasets, and 700 event types. The interpretation is conditional ('could') and grounded in the source's statement about limited applicability. It explicitly notes that the abstract does not establish performance in clinical or operational decisions. It does not turn 'competitive' into reliability.
Without task-specific fine-tuning, this framework achieves competitive zero-shot performance on eight unseen real-world datasets and scales up to 700 distinct event types.
media — pass: The selected media is the verified direct arXiv abstract page, the same URL as the source. It is the most direct way to access the paper.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
A large model trained on broad data that can be adapted to many different tasks.
Definition referenceDoing a task without worked examples (zero-shot) or with only a few examples given in the prompt (few-shot).
Definition referenceFurther training of an existing model to adapt it to a particular task or use.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivDoing a task without worked examples (zero-shot) or with only a few examples given in the prompt (few-shot).
Definition referenceA large model trained on broad data that can be adapted to many different tasks.
Definition referenceDoing a task without worked examples (zero-shot) or with only a few examples given in the prompt (few-shot).
Definition referenceFurther training of an existing model to adapt it to a particular task or use.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivDoing a task without worked examples (zero-shot) or with only a few examples given in the prompt (few-shot).
Definition reference