Create SynCo_modular_brain_agent_with_spikes_and_plasticity.py
Browse files
SynCo_modular_brain_agent_with_spikes_and_plasticity.py
ADDED
@@ -0,0 +1,252 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# MIT License
|
2 |
+
#
|
3 |
+
# Copyright (c) 2025 ALMUSAWIY Halliru
|
4 |
+
#
|
5 |
+
# Permission is hereby granted, free of charge, to any person obtaining a copy
|
6 |
+
# of this software and associated documentation files (the "Software"), to deal
|
7 |
+
# in the Software without restriction, including without limitation the rights
|
8 |
+
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
9 |
+
# copies of the Software, and to permit persons to whom the Software is
|
10 |
+
# furnished to do so, subject to the following conditions:
|
11 |
+
#
|
12 |
+
# The above copyright notice and this permission notice shall be included in all
|
13 |
+
# copies or substantial portions of the Software.
|
14 |
+
#
|
15 |
+
# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
16 |
+
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
17 |
+
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
18 |
+
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
19 |
+
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
20 |
+
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
|
21 |
+
# SOFTWARE.
|
22 |
+
|
23 |
+
# === V3 Modular Brain Agent with Plasticity - Block 1 ===
|
24 |
+
|
25 |
+
import torch
|
26 |
+
import torch.nn as nn
|
27 |
+
import torch.nn.functional as F
|
28 |
+
import numpy as np
|
29 |
+
import random
|
30 |
+
from torch.utils.data import DataLoader, Dataset
|
31 |
+
from collections import deque
|
32 |
+
from torchvision import datasets, transforms
|
33 |
+
|
34 |
+
# === Plastic Synapse Mechanisms ===
|
35 |
+
class PlasticLinear(nn.Module):
|
36 |
+
def __init__(self, in_features, out_features, plasticity_type="hebbian", learning_rate=0.01):
|
37 |
+
super().__init__()
|
38 |
+
self.in_features = in_features
|
39 |
+
self.out_features = out_features
|
40 |
+
self.weight = nn.Parameter(torch.randn(out_features, in_features) * 0.1)
|
41 |
+
self.bias = nn.Parameter(torch.zeros(out_features))
|
42 |
+
self.plasticity_type = plasticity_type
|
43 |
+
self.eta = learning_rate
|
44 |
+
self.trace = torch.zeros(out_features, in_features)
|
45 |
+
self.register_buffer('prev_y', torch.zeros(out_features))
|
46 |
+
|
47 |
+
def forward(self, x):
|
48 |
+
y = F.linear(x, self.weight, self.bias)
|
49 |
+
if self.training:
|
50 |
+
x_detached = x.detach()
|
51 |
+
y_detached = y.detach()
|
52 |
+
if self.plasticity_type == "hebbian":
|
53 |
+
hebb = torch.einsum('bi,bj->ij', y_detached, x_detached) / x.size(0)
|
54 |
+
self.trace = (1 - self.eta) * self.trace + self.eta * hebb
|
55 |
+
with torch.no_grad():
|
56 |
+
self.weight += self.trace
|
57 |
+
elif self.plasticity_type == "stdp":
|
58 |
+
dy = y_detached - self.prev_y
|
59 |
+
stdp = torch.einsum('bi,bj->ij', dy, x_detached) / x.size(0)
|
60 |
+
self.trace = (1 - self.eta) * self.trace + self.eta * stdp
|
61 |
+
with torch.no_grad():
|
62 |
+
self.weight += self.trace
|
63 |
+
self.prev_y = y_detached.clone()
|
64 |
+
return y
|
65 |
+
|
66 |
+
# === Spiking Surrogate Functions and Base Neurons ===
|
67 |
+
class SpikeFunction(torch.autograd.Function):
|
68 |
+
@staticmethod
|
69 |
+
def forward(ctx, input):
|
70 |
+
ctx.save_for_backward(input)
|
71 |
+
return (input > 0).float()
|
72 |
+
|
73 |
+
@staticmethod
|
74 |
+
def backward(ctx, grad_output):
|
75 |
+
input, = ctx.saved_tensors
|
76 |
+
return grad_output * (abs(input) < 1).float()
|
77 |
+
|
78 |
+
spike_fn = SpikeFunction.apply
|
79 |
+
|
80 |
+
class LIFNeuron(nn.Module):
|
81 |
+
def __init__(self, tau=2.0):
|
82 |
+
super().__init__()
|
83 |
+
self.tau = tau
|
84 |
+
self.mem = 0
|
85 |
+
|
86 |
+
def forward(self, x):
|
87 |
+
decay = torch.exp(torch.tensor(-1.0 / self.tau))
|
88 |
+
self.mem = self.mem * decay + x
|
89 |
+
out = spike_fn(self.mem - 1.0)
|
90 |
+
self.mem = self.mem * (1.0 - out.detach())
|
91 |
+
return out
|
92 |
+
|
93 |
+
# === Adaptive LIF Neuron ===
|
94 |
+
class AdaptiveLIF(nn.Module):
|
95 |
+
def __init__(self, size, tau=2.0, beta=0.2):
|
96 |
+
super().__init__()
|
97 |
+
self.size = size
|
98 |
+
self.tau = tau
|
99 |
+
self.beta = beta
|
100 |
+
self.mem = torch.zeros(size)
|
101 |
+
self.thresh = torch.ones(size)
|
102 |
+
|
103 |
+
def forward(self, x):
|
104 |
+
decay = torch.exp(torch.tensor(-1.0 / self.tau))
|
105 |
+
self.mem = self.mem * decay + x
|
106 |
+
out = spike_fn(self.mem - self.thresh)
|
107 |
+
self.thresh = self.thresh + self.beta * out
|
108 |
+
self.mem = self.mem * (1.0 - out.detach())
|
109 |
+
return out
|
110 |
+
|
111 |
+
# === Relay Layer with Attention ===
|
112 |
+
class RelayLayer(nn.Module):
|
113 |
+
def __init__(self, dim, heads=4):
|
114 |
+
super().__init__()
|
115 |
+
self.attn = nn.MultiheadAttention(embed_dim=dim, num_heads=heads, batch_first=True)
|
116 |
+
self.lif = LIFNeuron()
|
117 |
+
|
118 |
+
def forward(self, x):
|
119 |
+
attn_out, _ = self.attn(x, x, x)
|
120 |
+
return self.lif(attn_out)
|
121 |
+
|
122 |
+
# === Working Memory ===
|
123 |
+
class WorkingMemory(nn.Module):
|
124 |
+
def __init__(self, input_dim, hidden_dim):
|
125 |
+
super().__init__()
|
126 |
+
self.lstm = nn.LSTM(input_dim, hidden_dim, batch_first=True)
|
127 |
+
|
128 |
+
def forward(self, x):
|
129 |
+
out, _ = self.lstm(x)
|
130 |
+
return out[:, -1]
|
131 |
+
|
132 |
+
# === Place Cell Grid ===
|
133 |
+
class PlaceGrid(nn.Module):
|
134 |
+
def __init__(self, grid_size=10, embedding_dim=64):
|
135 |
+
super().__init__()
|
136 |
+
self.embedding = nn.Embedding(grid_size**2, embedding_dim)
|
137 |
+
|
138 |
+
def forward(self, index):
|
139 |
+
return self.embedding(index)
|
140 |
+
|
141 |
+
# === Mirror Comparator ===
|
142 |
+
class MirrorComparator(nn.Module):
|
143 |
+
def __init__(self, dim):
|
144 |
+
super().__init__()
|
145 |
+
self.cos = nn.CosineSimilarity(dim=1)
|
146 |
+
|
147 |
+
def forward(self, x1, x2):
|
148 |
+
return self.cos(x1, x2).unsqueeze(1)
|
149 |
+
|
150 |
+
# === Neuroendocrine Module ===
|
151 |
+
class NeuroendocrineModulator(nn.Module):
|
152 |
+
def __init__(self, input_dim, hidden_dim):
|
153 |
+
super().__init__()
|
154 |
+
self.lstm = nn.LSTM(input_dim, hidden_dim, batch_first=True)
|
155 |
+
|
156 |
+
def forward(self, x):
|
157 |
+
out, _ = self.lstm(x)
|
158 |
+
return out[:, -1]
|
159 |
+
|
160 |
+
# === Autonomic Feedback Module ===
|
161 |
+
class AutonomicFeedback(nn.Module):
|
162 |
+
def __init__(self, input_dim):
|
163 |
+
super().__init__()
|
164 |
+
self.feedback = nn.Linear(input_dim, input_dim)
|
165 |
+
|
166 |
+
def forward(self, x):
|
167 |
+
return torch.tanh(self.feedback(x))
|
168 |
+
|
169 |
+
# === Replay Buffer ===
|
170 |
+
class ReplayBuffer:
|
171 |
+
def __init__(self, capacity=1000):
|
172 |
+
self.buffer = deque(maxlen=capacity)
|
173 |
+
|
174 |
+
def add(self, inputs, labels, task):
|
175 |
+
self.buffer.append((inputs, labels, task))
|
176 |
+
|
177 |
+
def sample(self, batch_size):
|
178 |
+
indices = random.sample(range(len(self.buffer)), batch_size)
|
179 |
+
batch = [self.buffer[i] for i in indices]
|
180 |
+
inputs, labels, tasks = zip(*batch)
|
181 |
+
return inputs, labels, tasks
|
182 |
+
|
183 |
+
# === Full Modular Brain Agent with Plasticity ===
|
184 |
+
class ModularBrainAgent(nn.Module):
|
185 |
+
def __init__(self, input_dims, hidden_dim, output_dims):
|
186 |
+
super().__init__()
|
187 |
+
self.vision_encoder = nn.Linear(input_dims['vision'], hidden_dim)
|
188 |
+
self.language_encoder = nn.Linear(input_dims['language'], hidden_dim)
|
189 |
+
self.numeric_encoder = nn.Linear(input_dims['numeric'], hidden_dim)
|
190 |
+
|
191 |
+
# Plastic synapses (Hebbian and STDP)
|
192 |
+
self.connect_sensory_to_relay = PlasticLinear(hidden_dim * 3, hidden_dim, plasticity_type='hebbian')
|
193 |
+
self.relay_layer = RelayLayer(hidden_dim)
|
194 |
+
self.connect_relay_to_inter = PlasticLinear(hidden_dim, hidden_dim, plasticity_type='stdp')
|
195 |
+
|
196 |
+
self.interneuron = AdaptiveLIF(hidden_dim)
|
197 |
+
self.memory = WorkingMemory(hidden_dim, hidden_dim)
|
198 |
+
self.place = PlaceGrid(grid_size=10, embedding_dim=hidden_dim)
|
199 |
+
self.comparator = MirrorComparator(hidden_dim)
|
200 |
+
self.emotion = NeuroendocrineModulator(hidden_dim, hidden_dim)
|
201 |
+
self.feedback = AutonomicFeedback(hidden_dim)
|
202 |
+
|
203 |
+
self.task_heads = nn.ModuleDict({
|
204 |
+
task: nn.Linear(hidden_dim, out_dim)
|
205 |
+
for task, out_dim in output_dims.items()
|
206 |
+
})
|
207 |
+
|
208 |
+
self.replay = ReplayBuffer()
|
209 |
+
|
210 |
+
def forward(self, inputs, task, position_idx=None):
|
211 |
+
v = self.vision_encoder(inputs['vision'])
|
212 |
+
l = self.language_encoder(inputs['language'])
|
213 |
+
n = self.numeric_encoder(inputs['numeric'])
|
214 |
+
|
215 |
+
sensory_cat = torch.cat([v, l, n], dim=-1)
|
216 |
+
z = self.connect_sensory_to_relay(sensory_cat)
|
217 |
+
|
218 |
+
z = self.relay_layer(z.unsqueeze(1)).squeeze(1)
|
219 |
+
z = self.connect_relay_to_inter(z)
|
220 |
+
z = self.interneuron(z)
|
221 |
+
|
222 |
+
m = self.memory(z.unsqueeze(1))
|
223 |
+
p = self.place(position_idx if position_idx is not None else torch.tensor([0]))
|
224 |
+
e = self.emotion(z.unsqueeze(1))
|
225 |
+
f = self.feedback(z)
|
226 |
+
|
227 |
+
combined = z + m + p + e + f
|
228 |
+
out = self.task_heads[task](combined)
|
229 |
+
return out
|
230 |
+
|
231 |
+
def remember(self, inputs, labels, task):
|
232 |
+
self.replay.add(inputs, labels, task)
|
233 |
+
|
234 |
+
# === Main Test Block ===
|
235 |
+
if __name__ == "__main__":
|
236 |
+
input_dims = {'vision': 32, 'language': 16, 'numeric': 8}
|
237 |
+
output_dims = {'classification': 5, 'regression': 1, 'binary': 1}
|
238 |
+
agent = ModularBrainAgent(input_dims, hidden_dim=64, output_dims=output_dims)
|
239 |
+
|
240 |
+
tasks = list(output_dims.keys())
|
241 |
+
|
242 |
+
for step in range(250):
|
243 |
+
task = random.choice(tasks)
|
244 |
+
inputs = {
|
245 |
+
'vision': torch.randn(1, 32),
|
246 |
+
'language': torch.randn(1, 16),
|
247 |
+
'numeric': torch.randn(1, 8)
|
248 |
+
}
|
249 |
+
labels = torch.randint(0, output_dims[task], (1,)) if task == 'classification' else torch.randn(1, output_dims[task])
|
250 |
+
output = agent(inputs, task)
|
251 |
+
loss = F.cross_entropy(output, labels) if task == 'classification' else F.mse_loss(output, labels)
|
252 |
+
print(f"Step {step:02d} | Task: {task:13s} | Loss: {loss.item():.4f}")
|