Artificial Intelligence Network with 3D and Dual-Hemispheric Processing Inventor and Innovator: Nader Malik International Patent Registration Number: 140450140003002031June 9, 1404 PCT/IR2025/050026
Technical problems in the field of artificial intelligence neural networks in the inefficient processing of complex patterns, high cost of calculations, optimal energy consumption, memory consumption, more training time and interpretability are important problems in this field. Building neural networks from the functioning of the human brain is exactly based on the foundation of building artificial intelligence and this scientific implementation is close to simulating biological artificial intelligence. In the brain, data is sent as a signal in the thalamus to the two hemispheres of the brain with respect to amplification, filtering, classification and processing. And the two hemispheres with separate and coordinated tasks and division of complex patterns. For example: When reading a text, the left hemisphere processes words and sentences, and the right hemisphere helps to understand the overall meaning of the text and its relationship to the individual's previous knowledge. This is an undeniable scientific fact. In a 3D network with 3D processing, the next data signals in the first layer are pre-processed with the attention and modulation mechanism, amplified, filtered and de-noised, categorized and given in two weight vectors with separate outputs by the neurons projecting to the hidden layer which is given as a two-hemisphere architecture. The right output to the right hemisphere is nonlinear patterns in hyperbolic space and the left output to the left hemisphere is linear models which are separately and in coordination with the simulation of the corpus callosum in the neural network through the mechanisms of placement like an intermediate layer. From plasticity training, synaptic, reinforcement, deep and modulation signal processing methods are used to train the network. Using piezoelectric memory sensors and circuit bases that are deformed under the influence of environmental conditions (temperature, light, heat) and magnetic fields and controlled, monitored and programmed by the artificial intelligence coordinating module. Have high and efficient reception. Considering the submission of this abstract to the openAIba and Amazon competitions, this project is inspired and implemented by the eagle's optic nerve in observation and recognition.
Patent No. 140450140003002031
Inventor Nader Maleki
import os import glob import json import numpy as np import pandas as pd import matplotlib.pyplot as plt from matplotlib.backends.backend_pdf import PdfPages from sklearn.metrics import classification_report from datetime import datetime
plt.style.use('seaborn-v0_8')
---------- Part 1: Multi-run network processing ----------
results_dir = "results" os.makedirs(results_dir, exist_ok=True)
Reading files by listing once
all_files = glob.glob(os.path.join(results_dir, '*')) seed_metrics = sorted([f for f in all_files if 'metrics_seed' in f and f.endswith('.json')]) seed_histories = sorted([f for f in all_files if 'history_seed' in f and f.endswith('.csv')]) seed_confmats = sorted([f for f in all_files if 'confmat_seed' in f and f.endswith('.csv')])
Create PDF
pdf_path = os.path.join(results_dir, 'summary_dual_hemisphere_2031.pdf')
with PdfPages(pdf_path) as pdf:
Title page
fig, ax = plt.subplots(figsize=(11, 8.5)) ax.axis('off') ax.text(0.5, 0.6, "Dual Hemisphere Network Report – Code 2031", fontsize=20, ha='center', fontweight='bold') ax.text(0.5, 0.4, f"Generation date: {datetime.now().strftime('%Y-%m-%d %H:%M')}", fontsize=14, ha='center') pdf.savefig(fig) plt.close()
Process each run
for metrics_file, hist_file, conf_file in zip(seed_metrics, seed_histories, seed_confmats): seed_id = os.path.splitext(os.path.basename(metrics_file))[0].split("_")[-1]
Read data
with open(metrics_file, 'r', encoding='utf-8') as f: metrics = json.load(f)
history = pd.read_csv(hist_file) confmat = pd.read_csv(conf_file, index_col=0).values
Generate classification report
y_true = np.repeat(range(confmat.shape[0]), confmat.sum(axis=1)) y_pred = np.repeat(range(confmat.shape[1]), confmat.sum(axis=0))
report_text = ( f"Seed results {seed_id}:\n" f"Accuracy: {metrics.get('test_accuracy', 'N/A'):.4f}\n" f"F1 score: {metrics.get('f1_score', 'N/A'):.4f}\n" f"Classification report:\n{classification_report(y_true, y_pred)}" )
Training curves
fig, ax = plt.subplots(figsize=(8, 5)) ax.plot(history['epoch'], history['accuracy'], label='Training accuracy') ax.plot(history['epoch'], history['val_accuracy'], label='Accuracy Validation') ax.set_title(f'Learning Curves – Seed {seed_id}') ax.set_xlabel('Period') ax.set_ylabel('Accuracy') ax.legend() pdf.savefig(fig) plt.close()
Confusion Matrix
fig, ax = plt.subplots(figsize=(6, 5)) im = ax.imshow(confmat, cmap='viridis', aspect='auto') plt.colorbar(im, ax=ax) ax.set_title(f'Confusion Matrix – Seed {seed_id}') pdf.savefig(fig) plt.close()
Text Page
fig, ax = plt.subplots(figsize=(8.5, 11)) ax.axis('off') ax.text(0, 1, report_text, va='top', family='monospace', fontsize=8) pdf.savefig(fig) plt.close()
---------- Part 2: Piezoelectric Sensor Simulation ----------
Parameters (constant values)
PARAMS = { 'd33': 2.5e-10, # C/N 'thickness_m': 0.001, # m 'spring_k': 1000, # N/m 'damping_c': 10, # Ns/m 'capacitance_C': 1e-6, # Farad 'radius': 0.01 # m }
Time and Displacement
time = np.linspace(0, 1, 500) x = 0.01 * np.sin(2 * np.pi * 5 * time) dx_dt = np.gradient(x, time)
Calculations
cross_section = np.pi * PARAMS['radius'] ** 2 F_mech = PARAMS['spring_k'] * x + PARAMS['damping_c'] * dx_dt sigma = F_mech / cross_section Q = PARAMS['d33'] * F_mech V = Q / PARAMS['capacitance_C']
Plot graphs
fig, axs = plt.subplots(3, 1, figsize=(10, 12), sharex=True)
axs[0].plot(time, V * 1e3) axs[0].set_ylabel('Voltage (mV)') axs[0].set_title('Piezoelectric voltage vs time') axs[0].grid(True, alpha=0.3)
axs[1].plot(time, F_mech) axs[1].set_ylabel('Force (N)') axs[1].set_title('Mechanical force vs time') axs[1].grid(True, alpha=0.3)
axs[2].plot(time, sigma / 1e6) axs[2].set_ylabel('Pressure (MPa)') axs[2].set_xlabel('Time (s)') axs[2].set_title('Pressure vs time') axs[2].grid(True, alpha=0.3)
plt.tight_layout() pdf.savefig(fig) plt.close()
---------- Section 3: Innovations ----------
innovation_text = """ Model innovations:
- Combining a bi-hemispheric neural network with an array of intelligent piezoelectric sensors for simultaneous processing of mechanical and electrical data.
- Physical model including spring, damping and capacitance to simulate more accurately the behavior Sensor.
- Automatic multi-run reporting mechanism combining machine learning results and sensor circuits into a comprehensive PDF file. """
fig, ax = plt.subplots(figsize=(8.5, 11)) ax.axis('off') ax.text(0.05, 0.95, innovation_text, va='top', family='sans-serif', fontsize=12, linespacing=1. Final Report
Table of Results of Various Seed Experiments
| Seed | Experimental Accuracy | F1 Score |
|---|---|---|
| 1 | 0.9123 | 0.9087 |
| 2 | 0.9245 | 0.9201 |
| 3 | 0.9180 | 0.9142 |
| 4 | 0.9302 | 0.9265 |
| 5 | 0.9278 | 0.9239 |
Piezoelectric Sensor Simulation Summary Table
| Quantity | Peak Value | Unit |
|---|---|---|
| Voltage (V) | ±3.29 mV | mV |
| Force (F) | ±13.14 N | N |
| Pressure (σ) | ±0.042 MPa | MPa |
Analysis of Results
- Accuracy and F1 score remained above 0.90 in all five runs (Seeds), indicating the stability of the model.
Log in or sign up for Devpost to join the conversation.