MNIST reborn, restored and expanded. Now with an extra 50,000 training samples. If you used the original MNIST test set more than a few times, chances are your models overfit the test set. Time to test them on those extra samples. https://arxiv.org/abs/1905.10498

✴️ @AI_Python_EN

The field of statistics has very long history, dating back to ancient times.

Much of marketing data science can be traced to the origins of actuarial science, demography, sociology and psychology, with early statisticians playing major roles in all of these fields.

Big is relative, and statisticians have been working with "big data" all along. "Machine learners" such as SVM and random forests originated in statistics, and neural nets were inspired as much by regression as by theories of the human brain.

Statisticians are involved in a diverse range of fields, including marketing, psychology, pharmacology, economics, meteorology, political science and ecology, and have helped developed research methods and analytics for nearly any kind of data.

The history and richness of #statistics is not always appreciated, though. For example, this morning I was asked "How's your #machinelearning?" :-)

✴️ @AI_Python_EN

#NLP Podcast by Allen Institute for AI


✴️ @AI_Python_EN

💡💡 K-means Clustering – In-depth Tutorial with Example 💡💡

Credits - Data Flair

Link - https://lnkd.in/eCQUriR

#machineleaning #supervisedlearning #unsupervisedlearning #datascience #data #ai # #technology #deeplearning #artificalintelligence

✴️ @AI_Python_EN

Sampling is a deceptively complex subject, and some academic statisticians have devoted the bulk of their careers to it.

It's not a subject that thrills everyone but is a very important one, and one which seems underappreciated in marketing research and #data science.

Here are some books on or related to sampling I've found helpful:

- Survey Sampling (Kish)
- Sampling Techniques (Cochran)
- Model Assisted Survey Sampling (Särndal et al.)
- Sampling: Design and Analysis (Lohr)
- Practical Tools for Designing and Weighting Survey Samples (Valliant et al.)
- Survey Weights: A Step-by-step Guide to Calculation (Valliant and Dever)
- Complex Surveys (Lumley)
- Hard-to-Survey Populations (Tourangeau et al.)
- Small Area Estimation (Rao and Molina)

The first three are regarded as classics (though still relevant.) Sharon Lohr's book is the friendliest introduction I know of on this subject. Standard marketing research textbooks also give simple overviews of sampling but do not get into depth.

There are also academic journals that feature articles on sampling, such as the Public Opinion Quarterly (AAPOR) and the Journal of Survey #Statistics and Methodology (AAPOR and ASA).

✴️ @AI_Python_EN

Zero-shot Knowledge Transfer via Adversarial Belief Matching
Interesting paper on training student networks using a teacher network, without having access to the original training set.
Nowadays most pre-trained models are released without access to the actual data. The original data set might be sensitive in nature or very large. The idea here is to train a second network to learn the decision boundary of the large network.

Code:
http://bit.ly/2X9ChQt

Paper:
http://bit.ly/2XauCRS

✴️ @AI_Python_EN

Fact about #datascience practice in companies

✴️ @AI_Python_EN

This is Your Brain on Code 🧠💻🔢 computer programming is often associated with math, but researchers used functional MRI scans to show the role of the brain's language processing centers: https://lnkd.in/eN_-3RA

#datascience #machinelearning #ai #bigdata #analytics #statistics #artificialintelligence #datamining #computing #programmers #neuroscience

✴️ @AI_Python_EN

A nice explanation of backpropagation.
The notations are influenced by fast.ai (Deep Learning) program at USF and Deep Learning specialization course in Coursera.

https://lnkd.in/dthbv7U

#Deeplearning

✴️ @AI_Python_EN

#SparkNLP: State of the Art Natural Language Processing
Spark NLP ships with many NLP features, pre-trained models and pipelines #johnsnowlab

NLP Features:
#Tokenization; #Normalizer; #Stemmer; #Lemmatizer; #RegexMatching; #TextMatching; #Chunking; #DateMatcher; #Part-of-speech tagging; #SentenceDetector; #SentimentDetection (ML model); #SpellChecker (ML and DL models); #WordEmbeddings (#BERT and #GloVe); #Namedentityrecognition; #Dependencyparsing (Labeled/unlabled); Easy #TensorFlow integration; #pretrainedpipelines!

Github: https://lnkd.in/fbWquan
Website: https://lnkd.in/fRqsDHX

✴️ @AI_Python_EN

Simpson's paradox and Interpreting data

"A trend or result that is present when data is put into groups that reverses or disappears when the data is combined"

It is interesting to face these kind of challenges when working on the data and it gets even more interesting when you have to find way to select the right data points to make some concrete decisions.

Have a look at this article.

Link - https://lnkd.in/fnHswjM

I hope this helps! Have a productive weekend.

✴️ @AI_Python_EN

Cool paper written by Yoshua Bengio’s MILA team.

N-BEATS: Neural basis expansion analysis for interpretable time series forecasting.

Paper: arxiv.org/abs/1905.10437
GitHub for M4 dataset: https://lnkd.in/dWP5NmF
#timeseries #deeplearning

✴️ @AI_Python_EN

Human-level performance in first-person multiplayer games with population-based deep reinforcement learning

Jaderberg et al.: https://lnkd.in/eGTXYfE

#artificialintelligence #deeplearning #reinforcementlearning

✴️ @AI_Python_EN

#PyTorch Tensor To List: Convert a PyTorch Tensor To A Python List
http://bit.ly/2WfPPfK

✴️ @AI_Python_EN

Teaching language models grammar really does make them smarter

http://news.mit.edu/2019/teaching-language-models-grammar-makes-them-smarter-0529


✴️ @AI_Python_EN

Inferring a person’s looks from the way they speak: from a short input audio segment of a person speaking, the network directly reconstructs an image of the person’s face. Great: Clearly states ethical limits.
Paper: https://arxiv.org/pdf/1905.09773.pdf
Project: https://speech2face.github.io

✴️ @AI_Python_EN

Video: #AI, the end of #deeplearning ? #DeepLearning #QuantumMechanics #MachineLearning #CognitiveComputing #RPA #IntelligentAutomation

https://t.co/F5H5qfJeGk
✴️ @AI_Python_EN

check this collection of awesome deep Learning resources

https://github.com/frontbench-open-source/Data-Science-Free

✴️ @AI_Python_EN

THe best paper from ICLR 2019

https://www.analyticsvidhya.com/blog/2019/05/best-papers-iclr-2019/

Statistics, data science and jazz...

Though many jazz musicians have had extensive formal training in classical music - pianist Andre Previn being one notable example - some have been almost entirely self-taught - guitarist Wes Montgomery being one notable example.

Montgomery developed his own very unusual way of picking. If you're curious about what I mean, there are many videos on YouTube him performing.

He came from a musical family and he definitely had a talent for music. But the way he taught himself to play also had an influence on what he played, much of which was highly original, and his influence on jazz (and rock) continues to this day. He is generally regarded as one of the best guitarists ever.

What does this have to do with statistics and data science? Stats plays a vital role in data science, yet many data scientists are essentially self-taught or have learned from others who were largely self-taught.

I am not the only statistician concerned about misunderstandings and misuse of statistics in data science - Randy Bartlett, for one, has warned of a coming deluge of statistical malpractice. Some would argue that the deluge has arrived.

On the other hand, one wonders if data science will produce an equivalent of Wes Montgomery.

✴️ @AI_Python_EN

In a paper published recently, researchers from MIT’s Computer Science & Artificial Intelligence Laboratory have proposed a method for learning a face from audio recordings of that person speaking.

In their architecture, researchers utilize facial recognition pre-trained models as well as a face decoder model which takes as an input a latent vector and outputs an image with a reconstruction.

Paper: https://lnkd.in/fiUBjqh

#machinelearning #deeplearning #speech2face

✴️ @AI_Python_EN