EXD-Net v1: Hopfield Drift Memory Model

Emily D

Completed work

AI Model Developer

Data Modelling Analyst

Researcher

GitHub

Visual Studio Code

Artificial Intelligence

EXD-Net v1: Hopfield Drift Memory Model

This notebook models emotional memory destabilization and transformation using a classic Hopfield network. Memory scenes are encoded as binary patterns acting as attractors. Over time, new associations (e.g., "Benji") are reinforced, causing the network to drift, destabilize, and rewire.

The result is a computational simulation of:

identity shift

memory drift

nonlinear transformation

EXD-Net v1 simulates how emotional recall is altered by reinforcement and shows how some memories remain resistant, while others suddenly flip. While this first release focuses on emotional memory, EXD-Net is not limited to that domain. It is part of a broader framework designed to explore how instability, dissonance, and reinforcement reshape memory over time. By modeling memory scenes as attractor states and introducing disruptive associations, EXD-Net demonstrates how mental representations can shift, destabilize, and reorganize.

Core Concepts

Memory as Attractors: Each scene is a 32x32 binary pattern stored in a Hopfield network

Memory Drift: Reinforcing a new memory causes gradual or abrupt divergence from the original attractor

Destabilization Points: Some scenes show nonlinear turning points where recall shifts

Energy Landscapes: Hopfield energy is used to analyze the depth and dominance of emotional memory basins

Features

Binary image-based Hopfield network (32×32)

Progressive memory reinforcement with noise-robust recall

Cosine similarity tracking for association shift

Turning point detection (first recall of new association)

Hopfield energy comparison between old and new attractors

Convergence analysis over time

Interpretability through real emotions (e.g., denial, suppression, reattachment)

Repo Structure

.
├── exdnet_v1.ipynb        # Main notebook
├── data/
│   ├── exdnet_v1.csv      # Similarity logs
│   └── energy_df.csv      # Hopfield energy + turning points
├── images/                # Raw memory inputs (ex, benji, background)
├── plots/                 # Generated visualizations (delta plots, energy maps)
├── weights/
│   ├── W_base_v1.npy      # Weights trained only on Ex
│   └── W_temp_v1.npy      # Weights after Benji reinforcement
├── requirements.txt       # Minimal dependencies
└── README.md

Why It Matters

This model offers a computational lens on memory under instability. Not just emotional memory, but memory in flux. From simulating attachment rewiring to exploring cognitive dissonance, EXD-Net v1 shows how systems evolve through reinforcement, resistance, and collapse.

Drift isn't linear. Transformation isn't smooth. And not all memories are equal.

Next Steps

This is the foundational release before EXD-Net becomes a flexible framework for:

Modeling (emotional) memory as a dynamic, high-dimensional system

Simulating cognitive dissonance, destabilization trajectories, and transformation over time

Creating computational experiments rooted in cognitive and psychological theory

If you're interested in contributing, collaborating, or discussing aligned research directions, please reach out directly.

Read more about the theory behind EXD-Net on Substack: https://exd1000.substack.com

License: Custom non-commercial terms apply. Based on the Apache License 2.0 — reuse allowed with attribution. Commercial use or derivative frameworks require prior written permission from the author.

Like this project

Completed work

Posted May 31, 2025

Developed EXD-Net v1: a custom memory model that visualizes and quantifies memory drift via destabilization and adaptive learning over time.

Likes

Views