SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2207.02518 · NAACL · 2022

Compositional Generalization in Grounded Language Learning via Induced Model Sparsity

Alexander Ilin, Sam Spilsbury

arXiv · PDF · Open in the Atlas

Code that ran

7 of 10 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 7 have no contract verdict; 0 failures of our instrument; 3 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/2207.02518.json · badge: /api/ran/2207.02518.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
aalto-ai/sparse-compgen marked official in Papers with Code
Papers with Code archive
7 ran · 3 no recorded run
FunctionStatusWhere it lives
ImageBOWEmbedding Ran
ran on a synthesized input
aalto-ai/sparse-compgen/grounded_compgen_research/models/imitation/conv_transformer.py
code served (permissive licence) · get_code("ce9a198bf543d15d")
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, 32, 3, 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
import torch.nn as nn

def initialize_parameters(m):
    classname = m.__class__.__name__
    if classname.find('Linear') != -1:
        m.weight.data.normal_(0, 1)
        m.weight.data *= 1 / torch.sqrt(m.weight.data.pow(2).sum(1, keepdim=True))
        if m.bias is not None:
            m.bias.data.fill_(0)

class ImageBOWEmbedding(nn.Module):

    def __init__(self, max_value, n_channels, embedding_dim):
        super().__init__()
        self.max_value = max_value
        self.embedding_dim = embedding_dim
        self.embedding = nn.Embedding(n_channels * max_value, embedding_dim)
        self.apply(initialize_parameters)

    def forward(self, inputs):
        offsets = torch.Tensor([0, self.max_value, 2 * self.max_value, 3 * self.max_value]).to(inputs.device)
        offsetted = (inputs + offsets[None, None, None, :]).long()
        each_embedding = self.embedding(offsetted)
        each_embedding_flat = each_embedding.reshape(*each_embedding.shape[:-2], -1)
        return each_embedding_flat.permute(0, 3, 1, 2)


INIT_ARGS = {
    'max_value': 5,
    'n_channels': 4,
    'embedding_dim': 8,
}

ARGS = {
    'inputs': torch.randint(0, 5, (2, 3, 4, 4), dtype=torch.float32),
}

if __name__ == '__main__':
    model = ImageBOWEmbedding(**INIT_ARGS)
    output = model(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"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, 32, 3, 4) — shape (2, 32, 3, 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
ImageBlock Ran
ran on a synthesized input
aalto-ai/sparse-compgen/grounded_compgen_research/models/imitation/conv_transformer.py
code served (permissive licence) · get_code("f363048cbd07e14e")
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, 5, 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
import torch.nn as nn

def initialize_parameters(m):
    classname = m.__class__.__name__
    if classname.find('Linear') != -1:
        m.weight.data.normal_(0, 1)
        m.weight.data *= 1 / torch.sqrt(m.weight.data.pow(2).sum(1, keepdim=True))
        if m.bias is not None:
            m.bias.data.fill_(0)

class ImageBlock(nn.Module):

    def __init__(self, in_channels, nheads):
        super().__init__()
        self.attn = nn.MultiheadAttention(in_channels, nheads, dropout=0.1)
        self.bn1 = nn.LayerNorm(in_channels)
        self.mlp = nn.Sequential(nn.Linear(in_channels, in_channels), nn.ReLU(), nn.Linear(in_channels, in_channels), nn.ReLU())
        self.bn2 = nn.LayerNorm(in_channels)
        self.apply(initialize_parameters)

    def forward(self, x):
        batch, height, width, features = x.shape
        x = x.reshape(batch, height * width, features)
        x_permuted = x.permute(1, 0, 2)
        attended, _ = self.attn(x_permuted, x_permuted, x_permuted, need_weights=False)
        attended = attended.permute(1, 0, 2)
        attended_added = attended + x
        attended_added_norm = self.bn1(attended_added)
        attended_added_norm_mlp = self.mlp(attended_added_norm)
        attended_added_norm_mlp_norm = self.bn2(attended_added_norm_mlp)
        return attended_added_norm_mlp_norm.reshape(batch, height, width, features)


INIT_ARGS = {
    'in_channels': 4,
    'nheads': 2,
}

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

if __name__ == '__main__':
    model = ImageBlock(**INIT_ARGS)
    output = model(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Expected shape: (2, 3, 5, 4)")
    assert output.shape == (2, 3, 5, 4), f"Shape mismatch: {output.shape}"
    print("Success!")
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, 5, 4) — shape (2, 3, 5, 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
InstructionEncodingBlock Ran
ran on a synthesized input
aalto-ai/sparse-compgen/grounded_compgen_research/models/imitation/conv_transformer.py
code served (permissive licence) · get_code("0f60d4b3cc7f7bdb")
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 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 numpy as np
import torch
import torch.nn as nn

class PositionalEncoding(nn.Module):
    """This implementation is the same as in the Annotated transformer blog post
    See https://nlp.seas.harvard.edu/2018/04/03/attention.html for more detail.
    """

    def __init__(self, d_model, dropout=0.1, max_len=5000):
        assert d_model % 2 == 0, 'd_model should be an even number.'
        super(PositionalEncoding, self).__init__()
        pe = torch.zeros(max_len, d_model)
        position = torch.arange(0, max_len, dtype=torch.float).unsqueeze(1)
        div_term = torch.exp(torch.arange(0, d_model, 2).float() * (-np.log(10000.0) / d_model))
        pe[:, 0::2] = torch.sin(position * div_term)
        pe[:, 1::2] = torch.cos(position * div_term)
        pe = pe.unsqueeze(0).transpose(0, 1)
        self.register_buffer('pe', pe)

    def forward(self, x):
        return self.pe[:x.size(0), :]

class PositionalAddEncoding(nn.Module):

    def __init__(self, d_model, dropout=0.1, max_len=5000):
        super().__init__()
        self.pos_encoding = PositionalEncoding(d_model, max_len=max_len)
        self.dropout = nn.Dropout(p=dropout)

    def forward(self, x):
        pe = self.pos_encoding(x)
        return self.dropout(x + pe)

class InstructionEncodingBlock(nn.Module):

    def __init__(self, vocab_size, embedding_size, output_size):
        super().__init__()
        self.word_embedding = nn.Embedding(vocab_size, embedding_size)
        self.gru = nn.GRU(embedding_size, embedding_size)
        self.pos_encoding = PositionalAddEncoding(embedding_size)

    def forward(self, sentence):
        embedded = self.word_embedding(sentence.long())
        init_hidden = torch.zeros_like(embedded[:, 0, :].unsqueeze(0))
        embedded = embedded.permute(1, 0, 2)
        gru_output, _ = self.gru(embedded, init_hidden)
        pos_encoding = self.pos_encoding(gru_output)
        pos_encoding = pos_encoding.permute(1, 0, 2)
        gru_output = gru_output.permute(1, 0, 2)
        return (gru_output, pos_encoding)

INIT_ARGS = {
    'vocab_size': 100,
    'embedding_size': 4,
    'output_size': 8,
}

ARGS = {
    'sentence': torch.randint(0, 100, (2, 3)),
}

if __name__ == '__main__':
    model = InstructionEncodingBlock(**INIT_ARGS)
    result = model(**ARGS)
    print("Output shapes:")
    print(f"  gru_output: {result[0].shape}")
    print(f"  pos_encoding: {result[1].shape}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
numpy, 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
ObservationBlock Ran
ran on a synthesized input
aalto-ai/sparse-compgen/grounded_compgen_research/models/imitation/conv_transformer.py
code served (permissive licence) · get_code("6ff15de325f9d856")
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
import torch.nn as nn
import numpy as np

def initialize_parameters(m):
    classname = m.__class__.__name__
    if classname.find('Linear') != -1:
        m.weight.data.normal_(0, 1)
        m.weight.data *= 1 / torch.sqrt(m.weight.data.pow(2).sum(1, keepdim=True))
        if m.bias is not None:
            m.bias.data.fill_(0)

class ImageBOWEmbedding(nn.Module):

    def __init__(self, max_value, n_channels, embedding_dim):
        super().__init__()
        self.max_value = max_value
        self.embedding_dim = embedding_dim
        self.embedding = nn.Embedding(n_channels * max_value, embedding_dim)
        self.apply(initialize_parameters)

    def forward(self, inputs):
        offsets = torch.Tensor([0, self.max_value, 2 * self.max_value, 3 * self.max_value]).to(inputs.device)
        offsetted = (inputs + offsets[None, None, None, :]).long()
        each_embedding = self.embedding(offsetted)
        each_embedding_flat = each_embedding.reshape(*each_embedding.shape[:-2], -1)
        return each_embedding_flat.permute(0, 3, 1, 2)

class ImageBlock(nn.Module):

    def __init__(self, in_channels, nheads):
        super().__init__()
        self.attn = nn.MultiheadAttention(in_channels, nheads, dropout=0.1)
        self.bn1 = nn.LayerNorm(in_channels)
        self.mlp = nn.Sequential(nn.Linear(in_channels, in_channels), nn.ReLU(), nn.Linear(in_channels, in_channels), nn.ReLU())
        self.bn2 = nn.LayerNorm(in_channels)
        self.apply(initialize_parameters)

    def forward(self, x):
        batch, height, width, features = x.shape
        x = x.reshape(batch, height * width, features)
        x_permuted = x.permute(1, 0, 2)
        attended, _ = self.attn(x_permuted, x_permuted, x_permuted, need_weights=False)
        attended = attended.permute(1, 0, 2)
        attended_added = attended + x
        attended_added_norm = self.bn1(attended_added)
        attended_added_norm_mlp = self.mlp(attended_added_norm)
        attended_added_norm_mlp_norm = self.bn2(attended_added_norm_mlp)
        return attended_added_norm_mlp_norm.reshape(batch, height, width, features)

class ObservationBlock(nn.Module):

    def __init__(self, vocab_size, embedding_size, conv_dim, nheads):
        super().__init__()
        self.img_embeddings = ImageBOWEmbedding(16, 4, embedding_size)
        self.conv2x2 = nn.Conv2d(in_channels=embedding_size * 4, out_channels=conv_dim, kernel_size=2, padding=1)
        self.bn = nn.BatchNorm2d(conv_dim)
        self.relu = nn.ReLU(inplace=True)
        self.blocks = nn.Sequential(ImageBlock(conv_dim, nheads), ImageBlock(conv_dim, nheads), ImageBlock(conv_dim, nheads))

    def forward(self, img):
        embedded = self.img_embeddings(img)
        conv_embedded = self.conv2x2(embedded)
        conv_embedded_norm = self.bn(conv_embedded)
        conv_embedded_norm_relu = self.relu(conv_embedded_norm)
        conv_embedded_norm_relu = conv_embedded_norm_relu.permute(0, 2, 3, 1)
        return self.blocks(conv_embedded_norm_relu)

INIT_ARGS = {
    'vocab_size': 100,
    'embedding_size': 3,
    'conv_dim': 8,
    'nheads': 2
}

ARGS = {
    'img': torch.randint(0, 16, (2, 5, 7, 4))
}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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 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)
PositionalAddEncoding Ran
ran on a synthesized input
aalto-ai/sparse-compgen/grounded_compgen_research/models/imitation/conv_transformer.py
code served (permissive licence) · get_code("7a2b9fca729945a9")
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 3, dtype family float, shape (3, 2, 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 numpy as np
import torch
import torch.nn as nn

class PositionalEncoding(nn.Module):
    """This implementation is the same as in the Annotated transformer blog post
    See https://nlp.seas.harvard.edu/2018/04/03/attention.html for more detail.
    """

    def __init__(self, d_model, dropout=0.1, max_len=5000):
        assert d_model % 2 == 0, 'd_model should be an even number.'
        super(PositionalEncoding, self).__init__()
        pe = torch.zeros(max_len, d_model)
        position = torch.arange(0, max_len, dtype=torch.float).unsqueeze(1)
        div_term = torch.exp(torch.arange(0, d_model, 2).float() * (-np.log(10000.0) / d_model))
        pe[:, 0::2] = torch.sin(position * div_term)
        pe[:, 1::2] = torch.cos(position * div_term)
        pe = pe.unsqueeze(0).transpose(0, 1)
        self.register_buffer('pe', pe)

    def forward(self, x):
        return self.pe[:x.size(0), :]

class PositionalAddEncoding(nn.Module):

    def __init__(self, d_model, dropout=0.1, max_len=5000):
        super().__init__()
        self.pos_encoding = PositionalEncoding(d_model, max_len=max_len)
        self.dropout = nn.Dropout(p=dropout)

    def forward(self, x):
        pe = self.pos_encoding(x)
        return self.dropout(x + pe)


INIT_ARGS = {
    'd_model': 4,
    'dropout': 0.1,
    'max_len': 5000
}

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

if __name__ == '__main__':
    model = PositionalAddEncoding(**INIT_ARGS)
    output = model(**ARGS)
    print(f"Input shape: {ARGS['x'].shape}")
    print(f"Output shape: {output.shape}")
    print(f"Output:\n{output}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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 3, dtype family float, shape (3, 2, 4) — shape (3, 2, 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
PositionalEncoding Ran
ran on a synthesized input
aalto-ai/sparse-compgen/grounded_compgen_research/models/imitation/conv_transformer.py
code served (permissive licence) · get_code("6d88134a57f82380")
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 3, dtype family float, shape (7, 1, 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 numpy as np
import torch
import torch.nn as nn

class PositionalEncoding(nn.Module):
    """This implementation is the same as in the Annotated transformer blog post
    See https://nlp.seas.harvard.edu/2018/04/03/attention.html for more detail.
    """

    def __init__(self, d_model, dropout=0.1, max_len=5000):
        assert d_model % 2 == 0, 'd_model should be an even number.'
        super(PositionalEncoding, self).__init__()
        pe = torch.zeros(max_len, d_model)
        position = torch.arange(0, max_len, dtype=torch.float).unsqueeze(1)
        div_term = torch.exp(torch.arange(0, d_model, 2).float() * (-np.log(10000.0) / d_model))
        pe[:, 0::2] = torch.sin(position * div_term)
        pe[:, 1::2] = torch.cos(position * div_term)
        pe = pe.unsqueeze(0).transpose(0, 1)
        self.register_buffer('pe', pe)

    def forward(self, x):
        return self.pe[:x.size(0), :]


INIT_ARGS = {
    'd_model': 4,
    'dropout': 0.1,
    'max_len': 100,
}

ARGS = {
    'x': torch.randn(7, 2, 4),
}

if __name__ == '__main__':
    model = PositionalEncoding(**INIT_ARGS)
    output = model(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Output:\n{output}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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 3, dtype family float, shape (7, 1, 4) — shape (7, 1, 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
RelationalBlock Ran
ran on a synthesized input
aalto-ai/sparse-compgen/grounded_compgen_research/models/imitation/conv_transformer.py
code served (permissive licence) · get_code("11e6375c51d915d7")
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 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
import torch.nn as nn

# INIT_ARGS for RelationalBlock.__init__(sentence_embedding_dim, pos_encoding_dim, conv_dim, nheads)
INIT_ARGS = {
    'sentence_embedding_dim': 8,
    'pos_encoding_dim': 6,
    'conv_dim': 12,
    'nheads': 3
}

# ARGS for RelationalBlock.forward(x)
# x is a tuple of (encoded_image, encoded_sentence, positional_encodings)
batch = 2
height = 4
width = 5
features = 12  # must match conv_dim
sentence_len = 7

encoded_image = torch.randn(batch, height, width, features)
encoded_sentence = torch.randn(batch, sentence_len, features)
positional_encodings = torch.randn(batch, sentence_len, features)

ARGS = {
    'x': (encoded_image, encoded_sentence, positional_encodings)
}
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 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
ConvTransformerModel Not yet run
unverified (harvested, no recorded run)
aalto-ai/sparse-compgen/grounded_compgen_research/models/imitation/conv_transformer.py
code served (permissive licence) · get_code("86f07e914b4a1a28")
How it ran

no invocation recorded.

ProcessingTransformer Not yet run
unverified (harvested, no recorded run)
aalto-ai/sparse-compgen/grounded_compgen_research/models/imitation/conv_transformer.py
code served (permissive licence) · get_code("e6c51e69ff1b0120")
How it ran

no invocation recorded.

initialize_parameters Not yet run
unverified (harvested, no recorded run)
aalto-ai/sparse-compgen/grounded_compgen_research/models/imitation/conv_transformer.py
code served (permissive licence) · get_code("6125d291452ba72a")
How it ran

no invocation recorded.

Also present in other papers' code

Of 10 samples lifted from this paper's linked repositories, 10 carry a function identity and 0 do not. Of the 10 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 10 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

We provide a study of how induced model sparsity can help achieve compositional generalization and better sample efficiency in grounded language learning problems. We consider simple language-conditioned navigation problems in a grid world environment with disentangled observations. We show that standard neural architectures do not always yield compositional generalization. To address this, we design an agent that contains a goal identification module that encourages sparse correlations between words in the instruction and attributes of objects, composing them together to find the goal. 1 The output of the goal identification module is the input to a value iteration network planner. Our agent maintains a high level of performance on goals containing novel combinations of properties even when learning from a handful of demonstrations. We examine the

For agents

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

get_harvested_code_for_paper("2207.02518")
get_code_for_paper("2207.02518")
have("2207.02518")

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

curl https://syntology.ai/api/ran/2207.02518.json

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

[![Syntology run record](https://syntology.ai/api/ran/2207.02518.svg)](https://syntology.ai/paper/2207.02518)

Connect an agent — have() is free.