abstract_reasoning

l10n-placeholder1 หมายถึงชุดของประเภทความสัมพันธ์ (progression, XOR, OR, AND, union ที่สอดคล้องกัน), \\(O\\) หมายถึงประเภทวัตถุ (รูปร่าง เส้น) และ \\(A\\) หมายถึงประเภทแอตทริบิวต์ (ขนาด สี ตำแหน่ง, จำนวน). โครงสร้างของเมทริกซ์ \\(S\\)คือชุดของสาม \\(S={[r, o, a]}\\) ที่กำหนดความท้าทายที่เกิดจากเมทริกซ์เฉพาะ หากต้องการใช้ชุดข้อมูลนี้: ``` python import tensorflow_datasets as tfds ds = tfds.load('abstract_reasoning', split='train') for ex in ds.take(4): print(ex) ``` ดู [คู่มือ ](https://www.tensorflow.org/datasets/overview) สำหรับข้อมูลเพิ่มเติมเกี่ยวกับ [tensorflow_datasets](https://www.tensorflow.org/datasets) " />
  • คำอธิบาย :

ข้อมูล Procedurally Generated Matrices (PGM) จากเอกสาร Measuring Abstract Reasoning in Neural Networks, Barrett, Hill, Santoro และคณะ 2018 เป้าหมายคือการสรุปคำตอบที่ถูกต้องจากแผงบริบทตามเหตุผลเชิงนามธรรม

หากต้องการใช้ชุดข้อมูลนี้ โปรดดาวน์โหลดไฟล์ *.tar.gz ทั้งหมดจากหน้าชุดข้อมูลและวางไว้ใน ~/tensorflow_datasets/abstract_reasoning/

\(R\) หมายถึงชุดของประเภทความสัมพันธ์ (progression, XOR, OR, AND, union ที่สอดคล้องกัน), \(O\) หมายถึงประเภทวัตถุ (รูปร่าง เส้น) และ \(A\) หมายถึงประเภทแอตทริบิวต์ (ขนาด สี ตำแหน่ง ตัวเลข). โครงสร้างของเมทริกซ์\(S\)คือชุดของสามตัว \(S={[r, o, a]}\) ที่กำหนดความท้าทายที่เกิดจากเมทริกซ์เฉพาะ

แยก ตัวอย่าง
'test' 200,000
'train' 1,200,000
'validation' 20,000
  • โครงสร้างคุณลักษณะ :
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),
})
  • เอกสารคุณสมบัติ :
ลักษณะเฉพาะ ระดับ รูปร่าง Dประเภท คำอธิบาย
คุณสมบัติDict
คำตอบ วิดีโอ (รูปภาพ) (8, 160, 160, 1) uint8
บริบท วิดีโอ (รูปภาพ) (8, 160, 160, 1) uint8
ชื่อไฟล์ ข้อความ สตริง
meta_target เทนเซอร์ (12,) int64
relation_structure_encoded เทนเซอร์ (4, 12) int64
เป้า ป้ายกำกับคลาส int64
  • คีย์ภายใต้การดูแล (ดู as_supervised doc ): None

  • รูปภาพ ( tfds.show_examples ): ไม่รองรับ

  • การอ้างอิง :

@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 (การกำหนดค่าเริ่มต้น)

  • คำอธิบายการกำหนดค่า : โครงสร้างที่เข้ารหัสเมทริกซ์ในทั้ง
    ชุดการฝึกอบรมและการทดสอบประกอบด้วย \([r, o, a]\) สามเท่าสำหรับ \(r \\in R\),
    \(o \\in O\)และ \(a \\in A\)ชุดฝึกและชุดทดสอบไม่ปะติดปะต่อ
    การแยกที่เกิดขึ้นในระดับของตัวแปรอินพุต (เช่น พิกเซล
    อาการ).

  • ขนาดชุดข้อมูล : 42.02 GiB

  • ตัวอย่าง ( tfds.as_dataframe ):

abstract_reasoning/การแก้ไข

  • คำอธิบายการกำหนดค่า : ในการแบ่งกลาง \(S\) ประกอบด้วยใดๆ
    สามเท่า \([r, o, a]\). สำหรับการแก้ไขในชุดการฝึกเมื่อ
    แอตทริบิวต์คือ "สี" หรือ "ขนาด" (เช่น แอตทริบิวต์ที่สั่งซื้อ) ค่าของ
    แอตทริบิวต์ถูกจำกัดให้เป็นสมาชิกที่จัดทำดัชนีคู่ของชุดที่ไม่ต่อเนื่อง
    ในขณะที่ชุดทดสอบอนุญาตเฉพาะค่าดัชนีคี่เท่านั้น โปรดทราบว่าทั้งหมด
    \(S\) มี \([r, o, a]\) สามตัวที่มีแอตทริบิวต์สีหรือขนาด
    ดังนั้นคำถามทุกข้อในชุดทดสอบจึงจำเป็นต้องมีการสรุปทั่วไป

  • ขนาดชุดข้อมูล : 37.09 GiB

  • ตัวอย่าง ( tfds.as_dataframe ):

abstract_reasoning/การประมาณค่า

  • คำอธิบายการกำหนดค่า : เหมือนกับในการแก้ไข แต่ค่าของ
    แอตทริบิวต์ถูกจำกัดไว้ที่ครึ่งล่างของชุดแยกระหว่าง
    การฝึกอบรมในขณะที่ชุดการทดสอบใช้ค่าในครึ่งบน

  • ขนาดชุดข้อมูล : 35.91 GiB

  • ตัวอย่าง ( tfds.as_dataframe ):

abstract_reasoning/attr.rel.pairs

  • คำอธิบายการกำหนดค่า : \(S\) ทั้งหมดมีอย่างน้อยสองสามเท่า
    \(([r_1,o_1,a_1],[r_2,o_2,a_2]) = (t_1, t_2)\)ซึ่ง 400 สามารถทำงานได้ เรา
    สุ่มจัดสรร 360 ให้กับชุดฝึกและ 40 ให้กับชุดทดสอบ สมาชิก
    \((t_1, t_2)\) จาก 40 คู่ที่จัดไว้ไม่ได้เกิดขึ้นพร้อมกันในโครงสร้าง\(S\)
    ในชุดการฝึก และโครงสร้างทั้งหมด \(S\) มีคู่ดังกล่าวอย่างน้อยหนึ่งคู่
    \((t_1, t_2)\) เป็นส่วนย่อย

  • ขนาดชุดข้อมูล : 41.07 GiB

  • ตัวอย่าง ( tfds.as_dataframe ):

abstract_reasoning/attr.rels

  • คำอธิบายการกำหนดค่า : ในชุดข้อมูลของเรา มี 29 รายการที่เป็นไปได้ที่ไม่ซ้ำกัน
    สามเท่า \([r,o,a]\). เราจัดสรรเจ็ดสิ่งเหล่านี้สำหรับชุดทดสอบโดยการสุ่ม
    แต่คุณลักษณะแต่ละอย่างถูกแสดงเพียงครั้งเดียวในชุดนี้
    ทริปเปิลที่จัดขึ้นเหล่านี้ไม่เคยเกิดขึ้นกับคำถามในชุดการฝึกและ
    ทุก \(S\) ในชุดทดสอบมีอย่างน้อยหนึ่งตัว

  • ขนาดชุดข้อมูล : 41.45 GiB

  • ตัวอย่าง ( tfds.as_dataframe ):

abstract_reasoning/attrs.pairs

  • คำอธิบายการกำหนดค่า : \(S\) มีอย่างน้อยสองสามเท่า มี 20
    (ไม่เรียงลำดับ) คู่แอตทริบิวต์ที่ทำงานได้ \((a_1, a_2)\) เช่นนั้นสำหรับบางคน
    \(r_i, o_i, ([r_1,o_1,a_1],[r_2,o_2,a_2])\) เป็นคู่แฝดที่ทำงานได้
    \(([r_1,o_1,a_1],[r_2,o_2,a_2]) = (t_1, t_2)\). เราจัดสรร 16 คู่เหล่านี้
    สำหรับการฝึกอบรมและสี่สำหรับการทดสอบ สำหรับคู่ \((a_1, a_2)\) ในชุดทดสอบ
    \(S\) ในชุดการฝึกประกอบด้วย triple with \(a_1\) หรือ \(a_2\)ในการทดสอบ
    ชุด \(S\) ทั้งหมดมีสามเท่าด้วย \(a_1\) และ \(a_2\)

  • ขนาดชุดข้อมูล : 40.98 GiB

  • ตัวอย่าง ( tfds.as_dataframe ):

abstract_reasoning/attrs.shape.color

  • คำอธิบายการกำหนดค่า : แอตทริบิวต์ Held-out รูปร่าง-สี \(S\) นิ้ว
    ชุดฝึกอบรมไม่มีสามเท่าด้วย \(o\)=shape และ \(a\)=สี
    โครงสร้างทั้งหมดที่ควบคุมปริศนาในชุดทดสอบมีอย่างน้อยหนึ่งสาม
    ด้วย \(o\)=รูปร่าง และ \(a\)=สี

  • ขนาดชุดข้อมูล : 41.21 GiB

  • ตัวอย่าง ( tfds.as_dataframe ):

abstract_reasoning/attrs.line.type

  • คำอธิบายการกำหนดค่า : ประเภทบรรทัดแอตทริบิวต์ที่ถูกระงับ \(S\) นิ้ว
    ชุดฝึกอบรมไม่มีสามเท่าด้วย \(o\)=line และ \(a\)=type
    โครงสร้างทั้งหมดที่ควบคุมปริศนาในชุดทดสอบมีอย่างน้อยหนึ่งสาม
    ด้วย \(o\)=line และ \(a\)=type

  • ขนาดชุดข้อมูล : 41.40 GiB

  • ตัวอย่าง ( tfds.as_dataframe ):