SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2006.07123 · NeurIPS · 2020

Kernelized information bottleneck leads to biologically plausible 3-factor Hebbian learning in deep networks

Roman Pogodin, Peter Latham

arXiv · PDF · Open in the Atlas

Code that ran

4 of 4 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (2 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 2 have no contract verdict; 2 failures of our instrument; 0 no recorded run), measured 2026-11-22; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2006.07123.json · badge: /api/ran/2006.07123.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
romanpogodin/plausible-kernelized-bottleneck canonical (classifier label)
classifier basis: classifier_2025
4 ran · 0 no recorded run
FunctionStatusWhere it lives
HSICzyLoss Ran
ran on a synthesized input · constructed an object; no result was computed
romanpogodin/plausible-kernelized-bottleneck/losses.py
code served (permissive licence) · get_code("68c6a2eddcebbdf1")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, HSICzyLoss: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# Create a minimal Kernel class that implements the compute method
class MinimalKernel(nn.Module):
    def __init__(self, output_size):
        super().__init__()
        self.output_size = output_size
    
    def compute(self, x):
        """
        Compute kernel matrix.
        :param x: torch.Tensor of shape (batch_size, ...)
        :return: torch.Tensor of shape (batch_size, batch_size), symmetric kernel matrix
        """
        batch_size = x.shape[0]
        # Return a simple symmetric positive semi-definite matrix
        matrix = torch.randn(batch_size, batch_size)
        matrix = (matrix + matrix.t()) / 2  # Make symmetric
        matrix = matrix + torch.eye(batch_size) * 2  # Add to diagonal to ensure positive semi-definite
        return matrix

# Create kernel instances
z_kernel = MinimalKernel(output_size=4)
y_kernel = MinimalKernel(output_size=4)

# Create processing functions (optional)
z_processing = None
y_processing = None

# Define ARGS dictionary with all parameters for HSICzyLoss.__init__
ARGS = {
    'y_kernel': y_kernel,
    'z_kernel': z_kernel,
    'gamma': 2.0,
    'y_processing': y_processing,
    'z_processing': z_processing,
    'mode': 'biased'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type HSICzyLoss — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a HSICzyLoss output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
HSICzyLoss Ran
ran on a synthesized input · constructed an object; no result was computed
romanpogodin/plausible-kernelized-bottleneck/losses.py
code served (permissive licence) · get_code("65ed1e34a22d8274")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, HSICzyLoss: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# Define minimal Kernel classes that support .compute() method
class Kernel(nn.Module):
    def __init__(self):
        super().__init__()
    
    def compute(self, x):
        raise NotImplementedError

class LinearKernel(Kernel):
    def compute(self, x):
        # Compute linear kernel: K = x @ x.T
        batch_size = x.shape[0]
        x_flat = x.reshape(batch_size, -1)
        return torch.mm(x_flat, x_flat.t())

class GaussianKernel(Kernel):
    def __init__(self, sigma=1.0):
        super().__init__()
        self.sigma = sigma
    
    def compute(self, x):
        # Compute Gaussian RBF kernel
        batch_size = x.shape[0]
        x_flat = x.reshape(batch_size, -1)
        # Compute pairwise distances
        sq_norm = torch.sum(x_flat ** 2, dim=1, keepdim=True)
        pdist_sq = sq_norm + sq_norm.t() - 2.0 * torch.mm(x_flat, x_flat.t())
        pdist_sq = torch.clamp(pdist_sq, min=0.0)
        return torch.exp(-pdist_sq / (2.0 * self.sigma ** 2))

# Define simple processing functions
def flatten_processing(x):
    batch_size = x.shape[0]
    return x.reshape(batch_size, -1)

def normalize_processing(x):
    return x / (torch.norm(x, dim=1, keepdim=True) + 1e-8)

# Create kernel instances
z_kernel = LinearKernel()
y_kernel = GaussianKernel(sigma=1.0)

# Define processing pipelines (lists of callables)
z_processing = [flatten_processing, normalize_processing]
y_processing = [flatten_processing]

# Define ARGS dict with all parameters for HSICzyLoss.__init__
ARGS = {
    'y_kernel': y_kernel,
    'z_kernel': z_kernel,
    'gamma': 2.0,
    'y_processing': y_processing,
    'z_processing': z_processing,
    'mode': 'biased'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type HSICzyLoss — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a HSICzyLoss output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
estimate_hsic Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
romanpogodin/plausible-kernelized-bottleneck/losses.py
code served (permissive licence) · get_code("5b4431524c777f75")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned Tensor, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# Create two symmetric kernel matrices (Gram matrices)
# These represent k(a_i, a_j) and k(b_i, b_j)
batch_size = 2
dim = 4

# Create a random matrix and symmetrize it for a_matrix
a_random = torch.randn(batch_size, batch_size)
a_matrix = (a_random + a_random.T) / 2

# Create a random matrix and symmetrize it for b_matrix
b_random = torch.randn(batch_size, batch_size)
b_matrix = (b_random + b_random.T) / 2

ARGS = {
    'a_matrix': a_matrix,
    'b_matrix': b_matrix,
    'mode': 'biased'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
estimate_hsic_zy_objective Ran
ran; our contract draft was wrong, not the code
romanpogodin/plausible-kernelized-bottleneck/losses.py
code served (permissive licence) · get_code("2cce205bf9b0256f")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned Tensor, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# Define a simple Kernel class that computes kernel matrices
class SimpleKernel(nn.Module):
    def __init__(self, sigma=1.0):
        super().__init__()
        self.sigma = sigma
    
    def compute(self, x):
        """
        Compute RBF kernel matrix for input x.
        x: torch.Tensor of shape (batch_size, feature_dim)
        Returns: symmetric kernel matrix of shape (batch_size, batch_size)
        """
        # Compute pairwise squared Euclidean distances
        # ||x_i - x_j||^2 = ||x_i||^2 + ||x_j||^2 - 2*x_i*x_j
        x_norm = (x ** 2).sum(dim=1, keepdim=True)
        distances = x_norm + x_norm.t() - 2 * torch.mm(x, x.t())
        # RBF kernel: exp(-||x_i - x_j||^2 / (2*sigma^2))
        kernel_matrix = torch.exp(-distances / (2 * self.sigma ** 2))
        return kernel_matrix

# Create small tensors for z and y
batch_size = 2
z_dim = 4
y_dim = 3

z = torch.randn(batch_size, z_dim)
y = torch.randn(batch_size, y_dim)

# Create kernel instances
z_kernel = SimpleKernel(sigma=1.0)
y_kernel = SimpleKernel(sigma=1.0)

# Define arguments dictionary
ARGS = {
    'z': z,
    'y': y,
    'z_kernel': z_kernel,
    'y_kernel': y_kernel,
    'gamma': 2.0,
    'mode': 'biased'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block

Also present in other papers' code

Of 4 samples lifted from this paper's linked repositories, 4 carry a function identity and 0 do not. Of the 3 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 3 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

The state-of-the art machine learning approach to training deep neural networks, backpropagation, is implausible for real neural networks: neurons need to know their outgoing weights; training alternates between a bottom-up forward pass (computation) and a top-down backward pass (learning); and the algorithm often needs precise labels of many data points. Biologically plausible approximations to backpropagation, such as feedback alignment, solve the weight transport problem, but not the other two. Thus, fully biologically plausible learning rules have so far remained elusive. Here we present a family of learning rules that does not suffer from any of these problems. It is motivated by the information bottleneck principle (extended with kernel methods), in which networks learn to compress the input as much as possible without sacrificing prediction of the output. The resulting rules have a 3-factor Hebbian structure: they require pre-and post-synaptic firing rates and an error signal -the third factor -consisting of a global teaching signal and a layer-specific term, both available without a top-down pass. They do not require precise labels; instead, they rely on the similarity between pairs of desired outputs. Moreover, to obtain good performance on hard problems and retain biological plausibility, our rules need divisive normalization -a known feature of biological networks. Finally, simulations show that our rules perform nearly as well as backpropagation on image classification tasks.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2006.07123")
get_code_for_paper("2006.07123")
have("2006.07123")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2006.07123.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2006.07123.svg)](https://syntology.ai/paper/2006.07123)

Connect an agent — have() is free.