MNIST
A widely used benchmark dataset for handwritten digit recognition, containing 70,000 grayscale images of 28×28 pixels across 10 digit classes (0–9). It serves as a standard test set for introductory machine learning tasks and algorithm validation.
Core Facts
The MNIST dataset contains 60,000 training images and 10,000 test images, each a 28×28 pixel grayscale image of a handwritten digit.
On the MNIST dataset, a simple fully connected network can achieve over 98% accuracy, and a convolutional network can easily exceed 99.5%.
Timeline (last 90 days)
The experiments were tested only on MLPs, using only the MNIST and FashionMNIST datasets, with short training times and only one width tested, and have not yet involved CNN or Transformer architectures.
In the secondary test, SPreCap-e03 with χ1 set to 1.03 achieved a 5.28% test error rate on 100-layer MNIST, versus 6.36% for SELU, with a Holm-corrected p-value of 0.0064 and 13 of 15 seeds performing better.
In the task of implementing a CNN trained on MNIST using only NumPy and visualizing it, all three models (Kimi K3.1 Pro, Opus 5.5, DeepSeek V4.1 Flash) successfully completed the task.
The MNIST and CIFAR-10 datasets have become so common in tutorials that they lack discriminative value.
MNIST is often called the "Hello World" of deep learning.
In AI new-graduate resume screening, introductory projects like MNIST are typically regarded as "practice" rather than "work," and their signal value is far lower than projects involving real scenarios, complex pipelines, or deployment.
The experiment uses the EMNIST letters and MNIST digits datasets, constructing a distribution shift that is disjoint from the support of the original distribution yet completely blinds the CTM by conditionally resampling the EMNIST letters so that their conformal scores exactly match those of the MNIST digits.
Through empirical studies on different architectures (such as fully connected networks, convolutional networks, and residual networks) and datasets (such as MNIST, CIFAR-10, and ImageNet subsets), the authors found that the training process continuously expands ray coverage, especially in directions where the Jacobian singular values are smaller.
The MYOPIA benchmark contains four decision trees generated by LLMs and refined manually, corresponding to the GTSRB, MNIST, CIFAR-10, and MSCOCO datasets respectively.
The authors tested a variety of vision architectures ranging from D=7,850 to D=4,203,522 (about 4.2 million parameters) on the full MNIST benchmark.
21 more timeline events
All Facts (20)
The MNIST dataset contains 60,000 training images and 10,000 test images, each a 28×28 pixel grayscale image of a handwritten digit.
65%VerifiedOn the MNIST dataset, a simple fully connected network can achieve over 98% accuracy, and a convolutional network can easily exceed 99.5%.
65%Unverified招聘经理每天看到大量以'泰坦尼克号生存预测'或'MNIST手写识别'为主题的简历，这类教程项目已无法证明能力
70%UnverifiedThe experiments were tested only on MLPs, using only the MNIST and FashionMNIST datasets, with short training times and only one width tested, and have not yet involved CNN or Transformer architectures.
50%UnverifiedIn the secondary test, SPreCap-e03 with χ1 set to 1.03 achieved a 5.28% test error rate on 100-layer MNIST, versus 6.36% for SELU, with a Holm-corrected p-value of 0.0064 and 13 of 15 seeds performing better.
50%UnverifiedThe MNIST and CIFAR-10 datasets have become so common in tutorials that they lack discriminative value.
50%UnverifiedIn AI new-graduate resume screening, introductory projects like MNIST are typically regarded as "practice" rather than "work," and their signal value is far lower than projects involving real scenarios, complex pipelines, or deployment.
50%UnverifiedMNIST is often called the "Hello World" of deep learning.
50%UnverifiedThrough empirical studies on different architectures (such as fully connected networks, convolutional networks, and residual networks) and datasets (such as MNIST, CIFAR-10, and ImageNet subsets), the authors found that the training process continuously expands ray coverage, especially in directions where the Jacobian singular values are smaller.
50%UnverifiedThe MYOPIA benchmark contains four decision trees generated by LLMs and refined manually, corresponding to the GTSRB, MNIST, CIFAR-10, and MSCOCO datasets respectively.
50%UnverifiedThe current validation is still limited to MNIST and relatively small vision architectures, and whether the method can be extended to more complex architectures such as Transformers still requires further exploration.
50%UnverifiedIn the experiments, the classical weight-space bound rapidly inflated as the number of parameters increased, exceeding 940% on the largest model and completely failing.
50%UnverifiedThe data space DP-PAC-Bayes certificate remained stable across all model scales, with certified values in the range of 8.8%–9.4%.
50%UnverifiedALD's generation quality is validated on MNIST and CelebA64, and it is used for the data augmentation task on CIFAR-10LT.
50%UnverifiedImplementing a trainable Transformer or a hand-written MLP in NumPy on a CPU and getting it to run through MNIST is feasible, and many teaching projects do exactly this.
50%UnverifiedThe latest research, through full reproduction experiments, found that there is no predictable relationship between the training set size and the circuit size at which Q-FLAIR finally converges.
50%UnverifiedThe task of the reproduction experiments is binary classification of 3 versus 5 on the full-resolution (784-pixel) MNIST dataset, with the training set size N set at five levels from 2000 to 10000.
50%UnverifiedOn the MNIST handwritten digits and Oxford-IIIT Pet datasets, DI-SCUSUM exhibits lower conditional detection delay than SCUSUM at comparable false alarm levels.
50%UnverifiedOn the MNIST CNN task, MaRN compressed trainable parameters from 107,998 to 1,872 (approximately 57.7x), while maintaining 91.80% accuracy.
50%UnverifiedThe handwritten MLP model can achieve approximately 98.5% accuracy on the full MNIST training set.
50%