추상_추론

l10n-placeholder1은 일련의 관계 유형(진행, XOR, OR, AND, 일관된 합집합)을 나타내고 \\(O\\) 는 객체 유형(모양, 선)을 나타내고 \\(A\\) 은 속성 유형(크기, 색상, 위치, 번호). 행렬의 구조 \\(S\\)는 특정 행렬이 제기하는 문제를 결정하는 트리플 \\(S={[r, o, a]}\\) 집합입니다. 이 데이터셋을 사용하려면: ```python import tensorflow_datasets as tfds ds = tfds.load('abstract_reasoning', split='train') for ex in ds.take(4): print(ex) ``` [가이드 참조] ](https://www.tensorflow.org/datasets/overview)에서 [tensorflow_datasets](https://www.tensorflow.org/datasets)에 대한 자세한 내용을 확인하세요. " />
  • 설명 :

신경망 측정 추상 추론, Barrett, Hill, Santoro 외 논문의 PGM(Procedurally Generated Matrix) 데이터. 2018. 목표는 추상적인 추론을 기반으로 컨텍스트 패널에서 정답을 유추하는 것입니다.

이 데이터 세트를 사용하려면 데이터 세트 페이지에서 모든 *.tar.gz 파일을 다운로드하여 ~/tensorflow_datasets/abstract_reasoning/에 배치하십시오.

\(R\) 은 일련의 관계 유형(진행, XOR, OR, AND, 일관된 합집합)을 나타내고, \(O\) 은 객체 유형(모양, 선)을 나타내고, \(A\) 은 속성 유형(크기, 색상, 위치, 숫자). 행렬\(S\)의 구조는 특정 행렬이 제기하는 문제를 결정하는 트리플 \(S={[r, o, a]}\) 집합입니다.

나뉘다
'test' 200,000
'train' 1,200,000
'validation' 20,000
  • 기능 구조 :
FeaturesDict({
    'answers': Video(Image(shape=(160, 160, 1), dtype=uint8)),
    'context': Video(Image(shape=(160, 160, 1), dtype=uint8)),
    'filename': Text(shape=(), dtype=string),
    'meta_target': Tensor(shape=(12,), dtype=int64),
    'relation_structure_encoded': Tensor(shape=(4, 12), dtype=int64),
    'target': ClassLabel(shape=(), dtype=int64, num_classes=8),
})
  • 기능 문서 :
특징 수업 모양 D타입 설명
풍모Dict
답변 동영상(이미지) (8, 160, 160, 1) uint8
문맥 동영상(이미지) (8, 160, 160, 1) uint8
파일 이름 텍스트
meta_target 텐서 (12,) int64
related_structure_encoded 텐서 (4, 12) int64
표적 클래스 레이블 int64
@InProceedings{pmlr-v80-barrett18a,
  title =    {Measuring abstract reasoning in neural networks},
  author =   {Barrett, David and Hill, Felix and Santoro, Adam and Morcos, Ari and Lillicrap, Timothy},
  booktitle =    {Proceedings of the 35th International Conference on Machine Learning},
  pages =    {511--520},
  year =     {2018},
  editor =   {Dy, Jennifer and Krause, Andreas},
  volume =   {80},
  series =   {Proceedings of Machine Learning Research},
  address =      {Stockholmsmassan, Stockholm Sweden},
  month =    {10--15 Jul},
  publisher =    {PMLR},
  pdf =      {http://proceedings.mlr.press/v80/barrett18a/barrett18a.pdf},
  url =      {http://proceedings.mlr.press/v80/barrett18a.html},
  abstract =     {Whether neural networks can learn abstract reasoning or whetherthey merely rely on superficial statistics is a topic of recent debate. Here, we propose a dataset and challenge designed to probe abstract reasoning, inspired by a well-known human IQ test. To succeed at this challenge, models must cope with various generalisation 'regimes' in which the training data and test questions differ in clearly-defined ways. We show that popular models such as ResNets perform poorly, even when the training and test sets differ only minimally, and we present a novel architecture, with structure designed to encourage reasoning, that does significantly better. When we vary the way in which the test questions and training data differ, we find that our model is notably proficient at certain forms of generalisation, but notably weak at others. We further show that the model's ability to generalise improves markedly if it is trained to predict symbolic explanations for its answers. Altogether, we introduce and explore ways to both measure and induce stronger abstract reasoning in neural networks. Our freely-available dataset should motivate further progress in this direction.}
}

abstract_reasoning/neutral(기본 구성)

  • 구성 설명 : 두 가지 모두에서 행렬을 인코딩하는 구조
    교육 및 테스트 세트에는 \([r, o, a]\) 에 대한 \(r \\in R\)placeholder11 트리플이 포함됩니다.
    \(o \\in O\)및 \(a \\in A\). 훈련 세트와 테스트 세트는 분리되어 있습니다.
    입력 변수 수준에서 발생하는 분리(예: 픽셀
    표현).

  • 데이터세트 크기 : 42.02 GiB

  • ( tfds.as_dataframe ):

추상 추론/보간

  • 구성 설명 : 중립 분할에서와 같이 \(S\) 는
    트리플 \([r, o, a]\). 보간을 위해 훈련 세트에서
    속성이 "색상" 또는 "크기"(즉, 순서가 지정된 속성)인 경우
    속성은 불연속 세트의 짝수 인덱싱된 구성원으로 제한되었습니다.
    반면 테스트 세트에서는 홀수 색인 값만 허용되었습니다. 모든
    \(S\) 에는 color 또는 size 속성이 있는 삼중 \([r, o, a]\) 이 포함되어 있습니다.
    따라서 테스트 세트의 모든 질문에 대해 일반화가 필요합니다.

  • 데이터세트 크기 : 37.09 GiB

  • ( tfds.as_dataframe ):

추상 추론/외삽법

  • Config description : interpolation과 동일하지만,
    속성은 불연속 세트의 하위 절반으로 제한되었습니다.
    반면 테스트 세트에서는 상위 절반의 값을 취했습니다.

  • 데이터세트 크기 : 35.91 GiB

  • ( tfds.as_dataframe ):

abstract_reasoning/attr.rel.pairs

  • 구성 설명 : 모든 \(S\) 에는 최소 2개의 트리플이 포함되어 있습니다.
    \(([r_1,o_1,a_1],[r_2,o_2,a_2]) = (t_1, t_2)\), 그 중 400개가 실행 가능합니다. 우리
    훈련 세트에 360개, 테스트 세트에 40개를 무작위로 할당했습니다. 회원
    40개의 보류 쌍 중 l10n-\((t_1, t_2)\) 이 구조\(S\) 에서 함께 발생하지 않았습니다.
    트레이닝 세트에서 모든 구조 \(S\) 에는 이러한 쌍이 적어도 하나 있습니다.
    \((t_1, t_2)\) 를 하위 집합으로.

  • 데이터세트 크기 : 41.07 GiB

  • ( tfds.as_dataframe ):

abstract_reasoning/attr.rels

  • 구성 설명 : 데이터 세트에는 29개의 고유한
    트리플 \([r,o,a]\). 테스트 세트에 이 중 7개를 무작위로 할당했습니다.
    그러나 각 속성은 이 집합에서 정확히 한 번만 표시됩니다.
    이러한 홀드 아웃 트리플은 훈련 세트의 질문에서 발생하지 않았으며
    테스트 세트의 모든 \(S\) 에는 적어도 하나가 포함되어 있습니다.

  • 데이터세트 크기 : 41.45 GiB

  • ( tfds.as_dataframe ):

abstract_reasoning/attrs.pairs

  • 구성 설명 : \(S\) 에는 최소 2개의 트리플이 포함되어 있습니다. 20개가 있습니다.
    (정렬되지 않은) 속성 \((a_1, a_2)\) 의 실행 가능한 쌍으로 일부
    \(r_i, o_i, ([r_1,o_1,a_1],[r_2,o_2,a_2])\) 는 실행 가능한 트리플 페어입니다.
    \(([r_1,o_1,a_1],[r_2,o_2,a_2]) = (t_1, t_2)\). 이 쌍 중 16개를 할당했습니다.
    훈련용, 테스트용 4개. 테스트 세트의 쌍 \((a_1, a_2)\) 의 경우,
    훈련 세트의 l10n-\(S\) 에는 \(a_1\) 또는 \(a_2\)가 있는 트리플이 포함되어 있습니다. 테스트에서
    모든 \(S\) 에는 \(a_1\) 및 \(a_2\)이 있는 트리플이 포함되어 있습니다.

  • 데이터세트 크기 : 40.98 GiB

  • ( tfds.as_dataframe ):

abstract_reasoning/attrs.shape.color

  • 구성 설명 : 보류 속성 모양-색상. \(S\) in
    훈련 세트에는 \(o\)=모양 및 \(a\)=색상이 있는 트리플이 포함되지 않았습니다.
    테스트 세트의 퍼즐을 관리하는 모든 구조에는 적어도 하나의 트리플이 포함되어 있습니다.
    \(o\)= 모양 및 \(a\)= 색상.

  • 데이터세트 크기 : 41.21 GiB

  • ( tfds.as_dataframe ):

abstract_reasoning/attrs.line.type

  • 구성 설명 : 보류된 속성 라인 유형. \(S\) in
    트레이닝 세트에는 \(o\)=line 및 \(a\)=type의 트리플이 포함되어 있지 않습니다.
    테스트 세트의 퍼즐을 관리하는 모든 구조에는 적어도 하나의 트리플이 포함되어 있습니다.
    \(o\)= 라인 및 \(a\)= 유형.

  • 데이터세트 크기 : 41.40 GiB

  • ( tfds.as_dataframe ):