SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2006.05468 · ICML · 2021

Variational Auto-Regressive Gaussian Processes for Continual Learning

Sanyam Kapoor, Theofanis Karaletsos, Thang Bui

arXiv · PDF · Open in the Atlas

Code that ran

10 of 11 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 3 have no contract verdict; 7 failures of our instrument; 1 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; 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/2006.05468.json · badge: /api/ran/2006.05468.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
uber-research/vargp marked official in Papers with Code
Papers with Code archive
10 ran · 1 no recorded run
FunctionStatusWhere it lives
DeepRBFKernel Ran
ran on a synthesized input
uber-research/vargp/var_gp/vargp.py
code served (permissive licence) · get_code("83a90dd71ef26edf")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 4, dtype family float, shape (2, 2, 3, 5) 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
import torch.distributions as dist
from torch.distributions.kl import kl_divergence

class RBFKernel(nn.Module):

    def __init__(self, in_size, prior_log_mean=None, prior_log_logvar=None, map_est=False):
        super().__init__()
        self.map_est = map_est
        log_init = torch.tensor(0.5).log() * torch.ones(in_size + 1) + 0.05 * torch.randn(in_size + 1)
        self.log_mean = nn.Parameter(log_init)
        self.log_logvar = nn.Parameter(-2 * torch.ones(in_size + 1))
        self.register_buffer('prior_log_mean', prior_log_mean if prior_log_mean is not None else torch.zeros_like(self.log_mean))
        self.register_buffer('prior_log_logvar', prior_log_logvar if prior_log_logvar is not None else torch.zeros_like(self.log_logvar))

    def compute(self, kern_samples, x, y=None):
        """
        Generic batch kernel evaluation. Send
        y = None for efficient re-use of computations
        in case x = y.

        Arguments:
            kern_samples: n_hypers x (in_size + 1)
            x: ...batches x M x D
            y: ...batches x N x D, if None, assumed equals x

        Returns:
            n_hypers x ...batches x M x N
        """
        n_hypers = kern_samples.size(0)
        kern_samples = kern_samples.view(n_hypers, 1, *[1] * len(x.shape[:-2]), -1)
        sigma = kern_samples[..., :-1].exp()
        gamma2 = (kern_samples[..., -1:] * 2.0).exp()
        sx = x.unsqueeze(0) / sigma
        xx = torch.einsum('...ji,...ki->...jk', sx, sx)
        if y is None:
            yy = xy = xx
        else:
            sy = y.unsqueeze(0) / sigma
            yy = torch.einsum('...ji,...ki->...jk', sy, sy)
            xy = torch.einsum('...ji,...ki->...jk', sx, sy)
        dnorm2 = -2.0 * xy + xx.diagonal(dim1=-2, dim2=-1).unsqueeze(-1) + yy.diagonal(dim1=-2, dim2=-1).unsqueeze(-2)
        return gamma2 * (-0.5 * dnorm2).exp()

    def compute_diag(self, kern_samples):
        gamma2 = (kern_samples[..., -1:] * 2.0).exp().unsqueeze(-2)
        return gamma2

    def sample_hypers(self, n_hypers):
        if self.map_est:
            return self.log_mean.unsqueeze(0)
        log_dist = dist.Normal(self.log_mean, self.log_logvar.exp().sqrt())
        log_hypers = log_dist.rsample(torch.Size([n_hypers]))
        return log_hypers

    def kl_hypers(self):
        if self.map_est:
            return torch.tensor(0.0, device=self.log_mean.device)
        var_dist = dist.Normal(self.log_mean, self.log_logvar.exp().sqrt())
        prior_dist = dist.Normal(self.prior_log_mean, self.prior_log_logvar.exp().sqrt())
        total_kl = kl_divergence(var_dist, prior_dist).sum(dim=0)
        return total_kl

class DeepRBFKernel(RBFKernel):

    def __init__(self, in_size, feature_size=64, **kwargs):
        super().__init__(feature_size, **kwargs)
        self.phi = nn.Sequential(nn.Linear(in_size, 256), nn.ReLU(), nn.Linear(256, 256), nn.ReLU(), nn.Linear(256, feature_size))

    def compute(self, kern_samples, x, y=None):
        x = self.phi(x)
        if y is not None:
            y = self.phi(y)
        return super().compute(kern_samples, x, y=y)


INIT_ARGS = {
    'in_size': 4,
    'feature_size': 8,
    'map_est': False,
}

ARGS = {
    'kern_samples': torch.randn(2, 9),
    'x': torch.randn(2, 3, 4),
    'y': torch.randn(2, 5, 4),
}

METHOD = 'compute'

if __name__ == '__main__':
    model = DeepRBFKernel(**INIT_ARGS)
    result = getattr(model, METHOD)(**ARGS)
    print(f"Result shape: {result.shape}")
    print(f"Result: {result}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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 4, dtype family float, shape (2, 2, 3, 5) — shape (2, 2, 3, 5) 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
MulticlassSoftmax Ran
ran on a synthesized input
uber-research/vargp/var_gp/vargp.py
code served (permissive licence) · get_code("7054eeeb5bbdcb29")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 4, dtype family float, shape (2, 3, 4, 2) 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
import torch.nn.functional as F

class MulticlassSoftmax(nn.Module):

    def __init__(self, n_f=1):
        super().__init__()
        self.n_f = n_f

    def forward(self, mu, var):
        """
    Arguments:
      mu: n_hypers x out_size x B
      var: n_hypers x out_size x B

    Returns:
      Predictions for function samples.

      y_f: n_hypers x n_f x out_size x B
    """
        n_hypers, out_size, B = mu.shape
        eps = torch.randn(n_hypers, self.n_f, out_size, B, device=mu.device)
        f_samples = mu.unsqueeze(1) + var.sqrt().unsqueeze(1) * eps
        y_f = F.log_softmax(f_samples, dim=-2)
        return y_f

    def loss(self, pred_mu, pred_var, y):
        """
    Arguments:
      pred_mu: n_hypers x out_size x B
      pred_var: n_hypers x out_size x B
      y: B

    Returns:
      Total loss scalar value.
    """
        pred_y = self(pred_mu, pred_var).permute(3, 2, 0, 1)
        target_y = y.view(-1, 1, 1).expand(-1, *pred_y.shape[-2:])
        lik_loss = F.nll_loss(pred_y, target_y, reduction='none').mean(dim=-1).mean(dim=-1).sum(dim=0)
        return lik_loss

    def predict(self, mu, var):
        """
    Arguments:
      mu: n_hypers x out_size x B
      var: n_hypers x out_size x B

    Returns
      output probabilities B x out_size
    """
        y_f = self(mu, var)
        y = y_f.view(-1, *mu.shape[-2:])
        probs = y.logsumexp(dim=0).exp() / y.size(0)
        return probs.T


INIT_ARGS = {
    'n_f': 3
}

ARGS = {
    'mu': torch.randn(2, 4, 2),
    'var': torch.abs(torch.randn(2, 4, 2)) + 0.1
}

METHOD = 'forward'

if __name__ == '__main__':
    model = MulticlassSoftmax(**INIT_ARGS)
    output = model(**ARGS)
    print("Output shape:", output.shape)
    print("Output:", output)
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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 4, dtype family float, shape (2, 3, 4, 2) — shape (2, 3, 4, 2) 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
RBFKernel Ran
ran on a synthesized input
uber-research/vargp/var_gp/vargp.py
code served (permissive licence) · get_code("bcc66d9ddfa88032")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 4, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim 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

INIT_ARGS = {
    "in_size": 4,
    "prior_log_mean": torch.tensor([0.1, -0.2, 0.3, -0.1, 0.0]),
    "prior_log_logvar": torch.tensor([-1.0, -1.5, -2.0, -2.5, -3.0]),
    "map_est": False,
}

# Create input tensors for computation
x = torch.randn(2, 3, 4)  # batch=2, M=3, D=4
y = torch.randn(2, 5, 4)  # batch=2, N=5, D=4

ARGS = {
    "kern_samples": torch.randn(2, 5),  # n_hypers=2, in_size+1=5
    "x": x,
    "y": y,
}

METHOD = "compute"
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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 4, dtype family float — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
cholesky Ran
ran; our contract draft was wrong, not the code
uber-research/vargp/var_gp/vargp.py
code served (permissive licence) · get_code("09e7ee9467bbb608")
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 3, dtype family float, shape (2, 4, 4) 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

torch.manual_seed(0)

# Create a small positive definite matrix for Cholesky decomposition
# M should be symmetric positive definite
# Using a simple approach: create a random matrix and multiply by its transpose
batch_size = 2
dim = 4
random_matrix = torch.randn(batch_size, dim, dim)
M = torch.bmm(random_matrix, random_matrix.transpose(-2, -1)) + torch.eye(dim).unsqueeze(0)

ARGS = {
    "M": M,
    "eps": 0.0001
}
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 3, dtype family float, shape (2, 4, 4) — shape (2, 4, 4) 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
gp_cond Ran
ran; our contract draft was wrong, not the code
uber-research/vargp/var_gp/vargp.py
code served (permissive licence) · get_code("3c32ec62ac281327")
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 tuple. 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

torch.manual_seed(0)

# Parameters for gp_cond function
# u: ... x M x 1
# Kzz: ... x M x M
# Kzx: ... x M x N
# Kxx: ... x N x N

M = 4  # inducing points dimension
N = 5  # data points dimension
batch_size = 2

# u: batch x M x 1
u = torch.randn(batch_size, M, 1)

# Kzz: batch x M x M (covariance matrix, must be positive definite)
Kzz_base = torch.randn(batch_size, M, M)
Kzz = torch.bmm(Kzz_base, Kzz_base.transpose(-2, -1)) + 0.1 * torch.eye(M).unsqueeze(0)

# Kzx: batch x M x N
Kzx = torch.randn(batch_size, M, N)

# Kxx: batch x N x N (covariance matrix, must be positive definite)
Kxx_base = torch.randn(batch_size, N, N)
Kxx = torch.bmm(Kxx_base, Kxx_base.transpose(-2, -1)) + 0.1 * torch.eye(N).unsqueeze(0)

# Optional pre-computed Cholesky decomposition
Lz = None

# Optional pre-computed triangular solve result
Lz_Kzx = None

ARGS = {
    'u': u,
    'Kzz': Kzz,
    'Kzx': Kzx,
    'Kxx': Kxx,
    'Lz': Lz,
    'Lz_Kzx': Lz_Kzx,
}
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 tuple — a tuple output has no array shape of its own; none is 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
linear_joint Ran
ran; our contract draft was wrong, not the code
uber-research/vargp/var_gp/vargp.py
code served (permissive licence) · get_code("5bbde606973fa2d6")
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 tuple. 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

torch.manual_seed(0)

# Dimensions:
# M = 4 (latent dimension)
# N = 3 (observation dimension)
# Batch size = 2

M = 4
N = 3
batch_size = 2

# m: ... x M x 1
m = torch.randn(batch_size, M, 1)

# S: ... x M x M (covariance, must be positive definite)
S_temp = torch.randn(batch_size, M, M)
S = S_temp @ S_temp.transpose(-2, -1) + 0.1 * torch.eye(M).unsqueeze(0)

# Kzz: ... x M x M (covariance, must be positive definite)
Kzz_temp = torch.randn(batch_size, M, M)
Kzz = Kzz_temp @ Kzz_temp.transpose(-2, -1) + 0.1 * torch.eye(M).unsqueeze(0)

# Kzx: ... x M x N
Kzx = torch.randn(batch_size, M, N)

# V: ... x N x N (covariance, must be positive definite)
V_temp = torch.randn(batch_size, N, N)
V = V_temp @ V_temp.transpose(-2, -1) + 0.1 * torch.eye(N).unsqueeze(0)

# b: ... x N x 1
b = torch.randn(batch_size, N, 1)

# cache: optional dict
cache = {}

ARGS = {
    "m": m,
    "S": S,
    "Kzx": Kzx,
    "Kzz": Kzz,
    "V": V,
    "b": b,
    "cache": cache,
}
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 tuple — a tuple output has no array shape of its own; none is 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
linear_marginal_diag Ran
ran; our contract draft was wrong, not the code
uber-research/vargp/var_gp/vargp.py
code served (permissive licence) · get_code("b2df6ebf5ba61d4d")
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 tuple, ndim 3, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim 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

torch.manual_seed(0)

# Parameters for linear_marginal_diag
# m: ... x M x 1
# S: ... x M x M
# Kzx: ... x M x N
# Kzz: ... x M x M
# Kxx_diag: ... x N

M = 4  # latent dimension
N = 3  # output dimension
batch_shape = (2,)  # batch dimensions

# m: batch x M x 1
m = torch.randn(*batch_shape, M, 1)

# S: batch x M x M (must be positive definite)
S_temp = torch.randn(*batch_shape, M, M)
S = torch.matmul(S_temp, S_temp.transpose(-2, -1)) + 0.1 * torch.eye(M)

# Kzz: batch x M x M (must be positive definite)
Kzz_temp = torch.randn(*batch_shape, M, M)
Kzz = torch.matmul(Kzz_temp, Kzz_temp.transpose(-2, -1)) + 0.1 * torch.eye(M)

# Kzx: batch x M x N
Kzx = torch.randn(*batch_shape, M, N)

# Kxx_diag: batch x N (diagonal elements, must be positive)
Kxx_diag = torch.abs(torch.randn(*batch_shape, N)) + 0.1

# cache parameter (optional dict)
cache = {}

ARGS = {
    "m": m,
    "S": S,
    "Kzz": Kzz,
    "Kzx": Kzx,
    "Kxx_diag": Kxx_diag,
    "cache": cache,
}
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 tuple, ndim 3, dtype family float — a tuple output has no array shape of its own; none is stated
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
mat2trilvec Ran
ran; our contract draft was wrong, not the code
uber-research/vargp/var_gp/vargp.py
code served (permissive licence) · get_code("c00b44c81cb07664")
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 2, dtype family float, shape (2, 10) 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

torch.manual_seed(0)

# Create a batch of lower triangular matrices
# K=2 (batch size), m=4 (matrix dimension)
K = 2
m = 4
mat = torch.randn(K, m, m)

ARGS = {
    "mat": mat
}
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 2, dtype family float, shape (2, 10) — shape (2, 10) 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
rev_cholesky Ran
ran; our contract draft was wrong, not the code
uber-research/vargp/var_gp/vargp.py
code served (permissive licence) · get_code("07328e40f0f0f5dd")
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 3, dtype family float, shape (2, 4, 4) 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

torch.manual_seed(0)

# L should be a lower triangular matrix or batch of lower triangular matrices
# The function computes M = LL^T using einsum
# Shape: (..., i, j) where the last two dimensions are the matrix dimensions

L = torch.randn(2, 4, 4)
# Make it lower triangular for a valid Cholesky decomposition
L = torch.tril(L)

ARGS = {
    'L': L
}
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 3, dtype family float, shape (2, 4, 4) — shape (2, 4, 4) 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
vec2tril Ran
ran; our contract draft was wrong, not the code
uber-research/vargp/var_gp/vargp.py
code served (permissive licence) · get_code("ecbc8bc8c4ea9420")
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 3, dtype family float, shape (2, 3, 3) 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.functional as F

torch.manual_seed(0)

# The function vec2tril takes a vector of shape K x ((m * (m + 1)) // 2)
# and converts it to lower triangular matrices of shape K x m x m.
# 
# We need to construct:
# 1. vec: a tensor of shape K x ((m * (m + 1)) // 2)
# 2. m: an integer (optional, can be None)

# Let's choose m = 3 (a small lower triangular matrix)
# For m=3, the number of elements in lower triangular part is (3 * 4) // 2 = 6
m_val = 3
num_elements = (m_val * (m_val + 1)) // 2  # = 6

# Create a batch of vectors
# K = 2 (batch size)
K = 2
vec = torch.randn(K, num_elements)

ARGS = {
    "vec": vec,
    "m": m_val
}
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 3, dtype family float, shape (2, 3, 3) — shape (2, 3, 3) 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
VARGP Not yet run
unverified (harvested, no recorded run)
uber-research/vargp/var_gp/vargp.py
code served (permissive licence) · get_code("c01a80178407fc13")
How it ran

no invocation recorded.

Also present in other papers' code

Of 11 samples lifted from this paper's linked repositories, 11 carry a function identity and 0 do not. Of the 11 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 11 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

Through sequential construction of posteriors on observing data online, Bayes' theorem provides a natural framework for continual learning. We develop Variational Auto-Regressive Gaussian Processes (VAR-GPs), a principled posterior updating mechanism to solve sequential tasks in continual learning. By relying on sparse inducing point approximations for scalable posteriors, we propose a novel auto-regressive variational distribution which reveals two fruitful connections to existing results in Bayesian inference, expectation propagation and orthogonal inducing points. Mean predictive entropy estimates show VAR-GPs prevent catastrophic forgetting, which is empirically supported by strong performance on modern continual learning benchmarks against competitive baselines. A thorough ablation study demonstrates the efficacy of our modeling choices.

For agents

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

get_harvested_code_for_paper("2006.05468")
get_code_for_paper("2006.05468")
have("2006.05468")

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

curl https://syntology.ai/api/ran/2006.05468.json

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

[![Syntology run record](https://syntology.ai/api/ran/2006.05468.svg)](https://syntology.ai/paper/2006.05468)

Connect an agent — have() is free.