DREAM Documentation

Dynamic Recall and Elastic Adaptive Memory — a PyTorch implementation of continuous-time RNN cells with surprise-driven plasticity and liquid time-constants.

What is DREAM?

DREAM is a neural architecture for online adaptation during inference. Unlike static models (LSTM, Transformer), DREAM modulates its plasticity and integration speeds based on prediction error.

Core Components

• Surprise-Driven Plasticity — Hebbian learning modulated by prediction error
• Liquid Time-Constants (LTC) — Adaptive integration speeds
• Fast Weights — Low-rank decomposition for efficient meta-learning
• Sleep Consolidation — Memory stabilization during low-surprise periods

Installation

pip install dreamnn

Quick Start

import torch
from dream import DREAMConfig, DREAMCell

config = DREAMConfig(
    input_dim=80,
    hidden_dim=256,
    rank=16,
)

cell = DREAMCell(config)
state = cell.init_state(batch_size=4)

x = torch.randn(4, 100, 80)  # (batch, time, features)
output, final_state = cell.forward_sequence(x, return_all=True)

Documentation

Resources

• GitHub Repository
• PyPI Package
• Technical Report

Version: 0.1.2 (March 2026) | License: MIT