Event-Driven Models

Dimiter Dobrev

arxiv: 1906.10740 · v1 · pith:UP2ZGYR3new · submitted 2019-06-24 · 💻 cs.LG · cs.AI

Event-Driven Models

Dimiter Dobrev This is my paper

Pith reviewed 2026-05-25 17:30 UTC · model grok-4.3

classification 💻 cs.LG cs.AI

keywords event-driven modelsreinforcement learningobject identificationMarkov decision processessustainabilitypredictabilityaction-driven modelsinput output flows

0 comments

The pith

An object is an event-driven model whose state changes only when specific events occur.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper claims that reinforcement learning agents must first discover and identify objects before they can extract meaning from input and output flows rather than treating the data as noise. It defines an object as an event-driven model, a generalization of the action-driven models found in Markov decision processes. In action-driven models the state updates at every step, but event-driven models update only upon particular events that can be infrequent. This property is said to make the state more predictable and the model more sustainable. A reader would care because the definition supplies a concrete way to turn raw flows into stable, interpretable structures inside an agent.

Core claim

In Reinforcement Learning we look for meaning in the flow of input/output information. If we do not find meaning, the information flow is not more than noise to us. Before we are able to find meaning, we should first learn how to discover and identify objects. An object is an event-driven model. These models are a generalization of action-driven models. In Markov Decision Process we have an action-driven model which changes its state at each step. The advantage of event-driven models is their greater sustainability as they change their states only upon the occurrence of particular events. These events may occur very rarely, therefore the state of the event-driven model is much more predictab

What carries the argument

Event-driven model: a model that changes state only upon the occurrence of particular events rather than at every step, offered as the definition of an object in reinforcement learning.

If this is right

Agents can discover meaning in input/output flows by identifying event-driven models as objects.
Event-driven models provide greater sustainability than action-driven models because state changes are limited to rare events.
The state of an event-driven model is more predictable than that of an action-driven model.
This definition supplies a direct generalization of Markov decision process models for environments in which most steps produce no relevant change.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Agents might segment experience into object interactions by detecting the rare events that trigger state updates.
Standard step-by-step models may prove inefficient in settings dominated by long periods of no change.
The same distinction between frequent and rare updates could be applied to other sequential decision problems outside reinforcement learning.

Load-bearing premise

Identifying objects via event-driven models enables agents to discover meaning in input/output flows.

What would settle it

An experiment in which an agent equipped with event-driven object models extracts stable, meaningful structure from an RL input/output stream while an otherwise identical agent using only action-driven models continues to treat the same stream as noise.

read the original abstract

In Reinforcement Learning we look for meaning in the flow of input/output information. If we do not find meaning, the information flow is not more than noise to us. Before we are able to find meaning, we should first learn how to discover and identify objects. What is an object? In this article we will demonstrate that an object is an event-driven model. These models are a generalization of action-driven models. In Markov Decision Process we have an action-driven model which changes its state at each step. The advantage of event-driven models is their greater sustainability as they change their states only upon the occurrence of particular events. These events may occur very rarely, therefore the state of the event-driven model is much more predictable.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

This is a brief conceptual note claiming objects in RL are event-driven models that update only on rare events, but it supplies no definitions, arguments, or evidence.

read the letter

The main thing to know is that the paper redefines an object as an event-driven model, presented as a generalization of action-driven MDPs where states change every step. The claimed benefit is greater sustainability and predictability because updates happen only on infrequent events, which the author says helps agents find meaning in input/output flows rather than treating them as noise. That observation about per-step updates in standard MDPs is a fair point worth noting for world-model work. Beyond that, the paper does little else. It offers no formal definition of event-driven models, no mapping from I/O to such models, no derivation of the advantages, and no comparison to existing ideas on event-based or sparse RL. The benefits are asserted by construction rather than shown. There are no citations, experiments, or even a worked example. The central claim that this structure enables discovery of meaning is stated but not argued. This reads like an early idea sketch rather than a developed result. Readers already thinking about non-standard MDP variants or object-centric representations might skim it for the basic reframing, but most people working in RL will find nothing actionable or falsifiable here. It does not rise to the level that would justify sending it out for peer review.

Referee Report

3 major / 0 minor

Summary. The paper claims that in reinforcement learning, discovering meaning in input/output flows requires first identifying objects, and that an object is an event-driven model. This is presented as a generalization of action-driven models in Markov Decision Processes, where the latter change state at every step while event-driven models change state only upon rare events, conferring greater sustainability and state predictability.

Significance. If the central identification were formally defined and supported, the framework might offer a conceptual shift for object discovery in RL. As written, however, the manuscript consists of unsupported assertions without definitions, derivations, examples, or evidence, so it does not advance the field.

major comments (3)

[Abstract] Abstract: The claim that 'an object is an event-driven model' is asserted without any definition of 'object,' 'event,' or 'event-driven model,' and without a mapping from I/O flows to such models.
[Abstract] Abstract: The asserted advantage ('greater sustainability' and 'much more predictable' states) is built into the definition itself ('change their states only upon the occurrence of particular events. These events may occur very rarely') rather than derived from independent properties, benchmarks, or formal argument.
[Abstract] Abstract: No argument, derivation, or evidence is supplied for the opening premise that identifying objects via event-driven models will enable agents to 'discover meaning in the flow of input/output information,' which is required to connect the definition to the stated RL problem.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments. The manuscript is a brief conceptual note outlining an idea for object discovery in RL. We agree that the current version lacks explicit definitions, formal arguments, and supporting derivations, and we will revise the manuscript to address these shortcomings by expanding the text with the requested elements.

read point-by-point responses

Referee: [Abstract] Abstract: The claim that 'an object is an event-driven model' is asserted without any definition of 'object,' 'event,' or 'event-driven model,' and without a mapping from I/O flows to such models.

Authors: The referee is correct that no explicit definitions or mapping are provided in the current text. In the revised manuscript we will add formal definitions for 'object', 'event', and 'event-driven model', followed by a description of how input/output flows are mapped onto these models within an RL setting. revision: yes
Referee: [Abstract] Abstract: The asserted advantage ('greater sustainability' and 'much more predictable' states) is built into the definition itself ('change their states only upon the occurrence of particular events. These events may occur very rarely') rather than derived from independent properties, benchmarks, or formal argument.

Authors: We acknowledge that the stated advantages follow directly from the phrasing of the definition rather than from an independent derivation or comparison. The revision will include a separate section deriving the sustainability and predictability properties from the event-driven structure, using a comparison to standard action-driven MDPs and at least one illustrative example. revision: yes
Referee: [Abstract] Abstract: No argument, derivation, or evidence is supplied for the opening premise that identifying objects via event-driven models will enable agents to 'discover meaning in the flow of input/output information,' which is required to connect the definition to the stated RL problem.

Authors: The referee correctly observes that the link between event-driven object identification and meaning discovery is asserted without supporting reasoning. We will expand the introduction to provide a step-by-step argument showing how the identification of event-driven models reduces the effective state space and thereby aids extraction of meaningful structure from I/O streams. revision: yes

Circularity Check

0 steps flagged

No circularity: definitional claim with no derivation chain or self-referential reduction

full rationale

The paper asserts without derivation that 'an object is an event-driven model' and states its sustainability advantage directly from the definitional property of changing state only on events. No equations, predictions, or load-bearing steps are present that reduce to inputs by construction. The text contains no self-citations, fitted parameters renamed as predictions, or uniqueness theorems. The central claim is an unsupported identification rather than a derivation that collapses into its own premises, so no circularity patterns apply.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 1 invented entities

The central claim rests on the unproven premise that objects must be identified before meaning can be extracted from information flows and that event rarity directly yields predictability; no free parameters or invented entities with independent evidence are introduced, but the definition itself functions as an ad-hoc modeling choice.

axioms (2)

domain assumption Objects exist as identifiable entities within the flow of input/output information in RL
Invoked in the first paragraph to motivate the need for object discovery before meaning extraction.
ad hoc to paper Event-driven models are sustainable because relevant events occur rarely
Stated directly in the abstract as the source of the predictability advantage without further justification.

invented entities (1)

event-driven model no independent evidence
purpose: To serve as the definition of an object in RL
New modeling construct introduced to generalize action-driven MDPs; no independent evidence or falsifiable prediction supplied.

pith-pipeline@v0.9.0 · 5630 in / 1499 out tokens · 43457 ms · 2026-05-25T17:30:41.117778+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

7 extracted references · 7 canonical work pages · 1 internal anchor

[1]

Reinforcement Learning: An Introduction

Richard Sutton, Andrew Barto (1998). Reinforcement Learning: An Introduction. MIT Press, Cambridge, MA (1998)

work page 1998
[2]

Fourteen Declarative Principles of Experience-Oriented Intelligence

Richard Sutton (2008). Fourteen Declarative Principles of Experience-Oriented Intelligence. www.incompleteideas.net/RLAIcourse2009/principles2.pdf

work page 2008
[3]

Johan Åström. (1965). Optimal control of Markov processes with incomplete state information. Journal of Mathematical Analysis and Applications. 10: 174–205

work page 1965
[4]

Apostolos Burnetas, Michael Katehakis. (1997). Optimal Adaptive Policies for Markov Decision Processes. Mathematics of Operations Research. 22 (1): 222

work page 1997
[5]

Information Theories and Applications

Dimiter Dobrev (2017). How does the AI understand what’s going on . International Journal “Information Theories and Applications”, Vol. 24, Number 4, 2017, pp.345-369

work page 2017
[6]

Minimal and Maximal models in Reinforcement Learning

Dimiter Dobrev (2018). Minimal and Maximal models in Reinforcement Learning. August 2018. viXra:1808.0589

work page arXiv 2018
[7]

The Definition of AI in Terms of Multi Agent Systems

Dimiter Dobrev (2008). The Definition of AI in Terms of Multi Agent Systems. December 2008. arXiv:1210.0887

work page internal anchor Pith review Pith/arXiv arXiv 2008

[1] [1]

Reinforcement Learning: An Introduction

Richard Sutton, Andrew Barto (1998). Reinforcement Learning: An Introduction. MIT Press, Cambridge, MA (1998)

work page 1998

[2] [2]

Fourteen Declarative Principles of Experience-Oriented Intelligence

Richard Sutton (2008). Fourteen Declarative Principles of Experience-Oriented Intelligence. www.incompleteideas.net/RLAIcourse2009/principles2.pdf

work page 2008

[3] [3]

Johan Åström. (1965). Optimal control of Markov processes with incomplete state information. Journal of Mathematical Analysis and Applications. 10: 174–205

work page 1965

[4] [4]

Apostolos Burnetas, Michael Katehakis. (1997). Optimal Adaptive Policies for Markov Decision Processes. Mathematics of Operations Research. 22 (1): 222

work page 1997

[5] [5]

Information Theories and Applications

Dimiter Dobrev (2017). How does the AI understand what’s going on . International Journal “Information Theories and Applications”, Vol. 24, Number 4, 2017, pp.345-369

work page 2017

[6] [6]

Minimal and Maximal models in Reinforcement Learning

Dimiter Dobrev (2018). Minimal and Maximal models in Reinforcement Learning. August 2018. viXra:1808.0589

work page arXiv 2018

[7] [7]

The Definition of AI in Terms of Multi Agent Systems

Dimiter Dobrev (2008). The Definition of AI in Terms of Multi Agent Systems. December 2008. arXiv:1210.0887

work page internal anchor Pith review Pith/arXiv arXiv 2008