[POST] data generation

[miguelgfierro]

Alchemy

• Alchemists attempted to purify, mature, and perfect certain materials.[2][4][5][n 1] Common aims were chrysopoeia, the transmutation of "base metals" (e.g., lead) into "noble metals" (particularly gold);[2] the creation of an elixir of immortality;[2] the creation of panaceas able to cure any disease
• The start of Western alchemy may generally be traced to ancient and Hellenistic Egypt, where the city of Alexandria was a center of alchemical knowledge, and retained its pre-eminence through most of the Greek and Roman periods.
• Alchemy was spread in India by Alexander the Great, and later to China, the Arabic world and Europe.
• The terms "chemia" and "alchemia" were used as synonyms in the early XVI century, and the differences between alchemy, chemistry and small-scale assaying and metallurgy were not as neat as in the present day. There were important overlaps between practitioners, and trying to classify them into alchemists, chemists and craftsmen is anachronisti
• Isaac Newton devoted considerably more of his writing to the study of alchemy (see Isaac Newton's occult studies) than he did to either optics or physics.
• The decline of European alchemy was brought about by the rise of modern science with its emphasis on rigorous quantitative experimentation and its disdain for "ancient wisdom".
• Although the seeds of these events were planted as early as the 17th century, alchemy still flourished for some two hundred years, and in fact may have reached its peak in the 18th century. As late as 1781 James Price claimed to have produced a powder that could transmute mercury into silver or gold.
• Beginning around 1720, a rigid distinction was drawn between "alchemy" and "chemistry" for the first time.[80][81] By the 1740s, "alchemy" was now restricted to the realm of gold making, leading to the popular belief that alchemists were charlatans, and the tradition itself nothing more than a fraud.[78][81] In order to protect the developing science of modern chemistry from the negative censure of which alchemy was being subjected, academic writers during the scientific Enlightenment attempted, for the sake of survival, to divorce and separate the "new" chemistry from the "old" practices of alchemy.

Sources:

• https://en.m.wikipedia.org/wiki/Alchemy

[miguelgfierro]

Data size

• Google’s Research Director Peter Norvig claiming that “We don’t have better algorithms. We just have more data.”. linked to the article on “The Unreasonable Effectiveness of Data”
• the performance of machine learning algorithms changes with increasing data size in the case of traditional machine learning [10] algorithms (regression, etc.) and in the case of deep learning [11]. Specifically, for traditional machine learning algorithms, performance grows according to a power law and then reaches a plateau. Regarding deep learning, there is significant ongoing research as to how performance scales with increasing data size [12]-[16], [18].

source:

• https://towardsdatascience.com/how-do-you-know-you-have-enough-training-data-ad9b1fd679ee
• The World’s Most Valuable Resource Is No Longer Oil, But Data, https://www.economist.com/leaders/2017/05/06/the-worlds-most-valuable-resource-is-no-longer-oil-but-data May 2017.
• Martinez, A. G., No, Data Is Not the New Oil, https://www.wired.com/story/no-data-is-not-the-new-oil/ February 2019.

[miguelgfierro]

Data augmentation in CV

• Data generation with GANs https://arxiv.org/abs/1711.04340
• AutoAugment for object detection, achieving SOTA on COCO validation set (50.7 mAP). Paper: https://arxiv.org/abs/1906.11172, Code: https://github.com/tensorflow/tpu/tree/master/models/official/detection
• https://github.com/aleju/imgaug image augmentation library
• SOTA of image generation https://paperswithcode.com/task/image-generation
• Unsupervised Data Augmentation or UDA is a semi-supervised learning method which achieves state-of-the-art results on a wide variety of language and vision tasks. https://github.com/google-research/uda

[miguelgfierro]

Data augmentation in NLP

• Microsoft Researchers Banko and Brill [2001]” Scaling to Very Very Large Corpora for Natural Language Disambiguation“. different algorithms perform virtually the same. however, adding more examples (words) to the training set monotonically increases the accuracy of the model.
• LeakGAN text generation https://arxiv.org/abs/1709.08624
• NLP easy augmentation techniques https://github.com/jasonwei20/eda_nlp performance gain is especially large for small datasets at around 2–3% and modest for larger sizes (~1%)
• Unsupervised Data Augmentation or UDA is a semi-supervised learning method which achieves state-of-the-art results on a wide variety of language and vision tasks. https://github.com/google-research/uda

[miguelgfierro]

Data augmentation in tabular data

• Data augmentation for time series forecasting papers: https://stats.stackexchange.com/questions/320952/data-augmentation-strategies-for-time-series-forecasting
• A tensorflow implementation of Informative Generative Adversarial Network (InfoGAN ) to one dimensional ( 1D ) time series data with a supervised loss function. https://github.com/jeanjerome/semisupervised_timeseries_infogan/blob/master/README.md
• Data augmentation using synthetic data for time series classification with deep residual networks https://github.com/hfawaz/aaltd18/blob/master/README.md
• data augmentation methods for wearable sensor data (time-series data) https://github.com/terryum/Data-Augmentation-For-Wearable-Sensor-Data/blob/master/README.md

[miguelgfierro]

https://miguelgfierro.com/blog/2019/revisiting-the-revisit-of-the-unreasonable-effectiveness-of-data/