MétaCan
Menu
Back to cohort

Training Environments for Reinforcement Learning Cybersecurity Agents

2024· article· en· W4403024005 on OpenAlexaff
Alexandre Légère, Li Li, François Rivest, Ranwa Al Mallah

Bibliographic record

Venuenot available
Typearticle
Languageen
FieldComputer Science
TopicAdvanced Malware Detection Techniques
Canadian institutionsDefence Research and Development CanadaRoyal Military College of Canada
Fundersnot available
KeywordsReinforcement learningComputer scienceComputer securityTraining (meteorology)ReinforcementHuman–computer interactionArtificial intelligenceEngineering

Abstract

fetched live from OpenAlex

The development and training of Reinforcement Learning (RL) agents that can be applied to solve cybersecurity related challenges is a growing field of research. However, creating environments to train these agents effectively and efficiently is a nontrivial task. As part of the Markov Decision Process (MDP) involved in training an RL agent, the model must take actions on an environment and observe the impact of these actions on the state of the environment. For RL agents training in the field of cybersecurity, the environment being acted upon is often a computer network. Two primary streams of training environment have been developed for such agents: network emulators and network simulators. Emulators employ Virtual Machines (VMs) to provide a high-fidelity environment for the agent's interactions but suffer in efficiency since the actions of the agent are executed in real-time on the emulated network. Simulators employ abstracted models of the network environment that can rapidly compute the effect of an action on the environment. Simulated environments can train RL agents in much less time but introduce a simulation-to-reality gap that may result in an agent trained in a simulated environment that performs poorly in real environments. This paper reviews current emulated and simulated network environments for RL training.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.

How this classification was reachedexpand

Full frame distilled prediction

Teacher imitation

Not calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.

metaresearch head score (Codex)0.000
metaresearch head score (Gemma)0.000
Version: codex-gemma-dda1882f352aValidation status: machine_predicted_unvalidated
Candidate categoriesnone
Consensus categoriesnone
DomainCandidate signal: none · Consensus signal: none
Study designCandidate signal: Other design · Consensus signal: none
GenreCandidate signal: Methods · Consensus signal: none
Teacher disagreement score0.835
Threshold uncertainty score0.378

Codex and Gemma teacher scores by category

CategoryCodexGemma
Metaresearch0.0000.000
Meta-epidemiology (narrow)0.0000.000
Meta-epidemiology (broad)0.0000.000
Bibliometrics0.0000.000
Science and technology studies0.0000.000
Scholarly communication0.0000.001
Open science0.0000.000
Research integrity0.0000.000
Insufficient payload (model declined to judge)0.0000.000

Machine scores (provisional)

The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.

Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.

Opus teacher head0.039
GPT teacher head0.301
Teacher spread0.261 · how far apart the two teachers sit on this one work
Validation statusscore_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it

Classification

machine, unvalidated

Machine predicted; a candidate call from one teacher head, not a consensus.

The models applied no category: nothing in the taxonomy fit this work.
Study designOther design
Domainnot available
GenreMethods

How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".

Quick stats

Citations0
Published2024
Admission routes1
Has abstractyes

Explore more

Same topicAdvanced Malware Detection TechniquesFrench-language works237,207