Soft actor-critic
Algorithms AIMech:SoftActorCritic DRAFT
Aliases: SAC
An off-policy actor-critic algorithm fitting a stochastic policy and soft value functions under a reward-plus-policy-entropy objective.
Identity Notes & Concept Scope
SAC specifies an off-policy maximum-entropy actor-critic procedure. It is not a synonym for all reinforcement learning, for PPO, or for every entropy-regularized algorithm.
Claims & Grounded Evidence
1 Claim
Mechanism
#mechanism
Experience is retained in a replay buffer; value and policy networks are updated from sampled batches, with the policy objective balancing action value against log policy probability.
⚠️ Curated Limitations & Negative Results
- The inspected 2018 method includes a state-value approximator; later variants are not automatically identical in implementation.
- Tabular soft-policy-iteration theory does not by itself certify convergence of arbitrary neural function approximators.
Artificial Intelligence Ontology (AIO) Grounding
Mapped AIO Classes
| AIO Target Class | Mapping Relation | Mapping Rationale |
|---|---|---|
AIO:ReinforcementLearning
Reinforcement Learning
|
related_match | SAC is a particular algorithm for policy learning, whereas the pinned AIO target is a broad machine-learning task. |
Evidence Sources & Primary Literature
1 Source| Source ID | Publication / Document | Type / Access | Reference & Checksum |
|---|---|---|---|
original-paper
|
Tuomas Haarnoja, Aurick Zhou, Pieter Abbeel, Sergey Levine
|
primary paper
full_text
|
ID: arXiv:1801.01290v2
sha256:5c33fae017d0…
|
Curation Provenance
Curator: Codex
Method: ai_assisted
Model: GPT-6
Created: 2026-10-11T03:12:53Z
Updated: 2026-10-11T03:12:53Z
Source File:
data/curated/SoftActorCritic.yaml