Have you heard of SuperTML?

Two-Dimensional Word Embedding and Transfer Learning Using ImageNet Pretrained CNN Models for the Classifications on Tabular Data


SuperTML: Two-Dimensional Word Embedding and Transfer Learning Using ImageNet Pretrained CNN Models for the Classifications on Tabular Data

Tabular data is the most commonly used form of data in industry. Gradient Boosting Trees, Support Vector Machine, Random Forest, and Logistic Regression are typically used for classification tasks on tabular data.

DNN models using categorical embeddings are also applied in this task, but all attempts thus far have used one-dimensional embeddings. The recent work of Super Characters method using two-dimensional word embeddings achieved the state of art result in text classification tasks, showcasing the promise of this new approach.

The SuperTML method, which borrows the idea of Super Characters method and two-dimensional embeddings to address the problem of classification on tabular data. It has achieved state-of-the-art results on both large and small datasets.


Here’s the paper: https://lnkd.in/djGFf63

❇️ @AI_Python
🗣 @AI_Python_arXiv
✴️ @AI_Python_EN

Data-Driven Careers Dechipered


Seems Different with what I believed, but still worth it to discuss

#business #technology #datascience

❇️ @AI_Python
🗣 @AI_Python_arXiv
✴️ @AI_Python_EN

Yann LeCun Tweeted:

I gave two talks in Harvard's Mind, Brain, & Behavior Distinguished Lecture Series last week.

The slides are here:
- 2019-03-13 "The Power and Limits of Deep Learning":...

https://t.co/Pp4flTGlZ8

✴️ @AI_Python_EN

✴️ @AI_Python_EN

Our Computer Vision and Deep Learning Group:

https://t.me/joinchat/ECtp7VVFvEwjIrMrdgI-2w

🗣 @AI_Python_Arxiv
✴️ @AI_Python_EN
❇️ @AI_Python

Many misunderstandings persist regarding logistic regression (LR).

Though it can be used for classification, it is not a classification method. Its predictions are model-based estimates of the probabilities of group/class membership.

Cutoffs can be drawn anywhere that are meaningful to decision makers, not just at probability = .50. Automatically using the .50 cutoff - the default for most LR programs - is a mistake, especially when group/class sizes are highly imbalanced.

Relationships between the predictors (independent variables) and outcome (dependent variable) do not have to be "linear" - any sort of relationship, including curvilinear and moderated relationships (interactions), can be modeled.

Binary LR is just one member of the GLM family. There are also LR models for multinomial, ordinal and count data, as well as probit analysis.

More advanced extensions include multilevel and mixture models, and SEM with multiple categorical outcomes. There are also Bayesian alternatives to maximum likelihood estimation.

BIG topic, with very practical implications for marketing research, data science and many other fields. I'm still learning about it.

🗣 @AI_Python_Arxiv
✴️ @AI_Python_EN
❇️ @AI_Python

A Brief History of Data Science (Pre-2010, i.e. prior to rise of deep learning & popular usage of the term "data science")
#
Note: Modified original version of infographic to add 3 seminal developments in the history of Artificial Intelligence:

- 1943: Artificial neuron model (McCulloch & Pitts)
- 1950: Turing Test (Alan Turing)
- 1956: Dartmouth Conference (McCarthy, Minsky, Shannon)

#datascience #statistics #analytics #machinelearning #bigdata #artificialintelligence #innovation #technology #history #ai #datamining #informatics #infographics #informationtechnology #computerscience #dataanalysis #deeplearning #neuroscience #mathematics #science

🗣 @AI_Python_Arxiv
✴️ @AI_Python_EN
❇️ @AI_Python

Comprehensive Collection of #DataScience and #MachineLearning Resources for #DataScientists includes “Great Articles on Natural Language Processing” +much more 👉https://bit.ly/2nvMXIx #abdsc #BigData #AI #DeepLearning #Databases #Coding #Python #Rstats #NeuralNetworks #NLProc

✴️ @AI_Python_EN

Coloring massive graphs is a notoriously difficult problem to solve…Deep reinforcement learning to the rescue! Jiayi Huang et al use a novel architecture (FastColorNet) to learn new state of the art heuristics for graph coloring: https://lnkd.in/gcE8cWz #TechRec

✴️ @AI_Python_EN

Statisticians often refer to observed variables and latent variables. Put simply, an observed variable is what we have in our data file. It may be useful as is but, in fields such as marketing research, may be an indicator of a latent variable, often called a factor, which we cannot directly observe. In a more familiar context, a high fever may be an indication of influenza, which our doctor cannot observe directly.

The distinction between the two has practical importance. For instance, if a survey respondent selects Convenience as one of the reason she shops at Chain X, it could indicate many things, for example, that the store is physically nearby, that parking is easy, that checkout is fast, or that the store layout makes it easy to find what she wants.

In questionnaire design and when analyzing survey data, marketing researchers frequently confuse observed and latent variables - "Effective" is another example. We tend to zero in on specific items or statements when we should be thinking about the constructs they represent. Also, the questions we ask may be confusing or meaningless to respondents because of this.

It's best to think about these issues when designing our research and planning our analytics.

✴️ @AI_Python_EN

How to Combine Neural Networks and Decision Trees

#NeuralNetworks #decisiontree

✴️ @AI_Python_EN

While #AI hype machine is in overdrive we are thrilled to see that #DeepLearning is making steady inroads into making lives of #opthalmologists easier.

We believe in this too and include practical and hands-on datasets in our own trainings so more engineers and doctors can work together to diagnose such diseases!

Diabetic retinopathy (DR), a major microvascular complication of diabetes, has a significant impact on the world's health systems. Globally, the number of people with DR will grow from 126.6 million in 2010 to 191.0 million by 2030.

In U.S alone, more than 29 million people have diabetes, and are at risk for diabetic retinopathy, a potentially blinding eye disease. People typically don't notice changes in their vision in the disease's early stages. But as it progresses, diabetic retinopathy usually causes vision loss that in many cases cannot be reversed. That's why it's so important that people with diabetes have yearly screenings.

Unfortunately, the accuracy of screenings can vary significantly. One study found a 49 percent error rate among internists, diabetologists, and medical residents. This is really bad!

Read Google's research https://lnkd.in/gDMW-fD
#artificialintelligence #diabeticretinopathy

✴️ @AI_Python_EN

Artificial intelligence: art’s weird and wonderful new medium

By Francesca Gavine, how to spend it: https://lnkd.in/eXEFkkQ

#art #artificialintelligence #deeplearning
#generativeadversarialnetworks

✴️ @AI_Python_EN

HOW DO I LEARN BASICS MACHINE LEARNING IN 30 DAYS ?


Harshit Ahluwalia updated his github about 30 days strategy to learn machine learning from basics. This one gonna very useful for coaching session on Internship program https://lnkd.in/fJaF-jj

#machinelearning #datascience #python #artificialintelligence #repository #100DaysOfMLCode

✴️ @AI_Python_EN

Access free GPU compute via Colab

By Google: https://lnkd.in/ds_j5nz

Colaboratory is a research tool for machine learning education and research. It’s a Jupyter notebook environment that requires no setup to use.

#ArtificialIntelligence #DeepLearning #MachineLearning

✴️ @AI_Python_EN

Reducing the Need for Labeled Data in Generative Adversarial Networks #DataScience #MachineLearning #ArtificialIntelligence

http://bit.ly/2FqeJiF

✴️ @AI_Python_EN

A PyTorch implementation of BigGAN with pretrained weights and conversion scripts

By Thomas Wolf: https://lnkd.in/e_Pph_T

#pytorch #biggan #computervision #artificialintelligence
#generativeadversarialnetwork

✴️ @AI_Python_EN

Interpretable machine learning is so important doesn’t matter whether you want to understand a simple linear regression model to more complex ones like neural networks. Understanding your models help to prevent biases, gain trust and help you building better models. If you haven’t done it yet then start with it now! It’s never too late! #deeplearning #machinelearning #explainableAI

✴️ @AI_Python_EN

I find Mark Berliner’s Bayesian Hierarchical Model (BHM) paradigm helpful.

At the top level is the data model, which is a probability model that specifies the distribution of the data given an underlying “true” process (sometimes called the hidden or latent process) and given some parameters that are needed to specify this distribution.

At the next level is the process model, which is a probability model that describes the hidden process (and, thus, its uncertainty) given some parameters. Note that at this level the model does not need to account for measurement uncertainty. The process model can then use science-based theoretical or empirical knowledge, which is often physical or mechanistic.

At the bottom level is the parameter model, where uncertainty about the parameters is modeled. From top to bottom, the levels of a BHM are:

1. Data model: [data|process, parameters]
2. Process model: [process|parameters]
3. Parameter model: [parameters]

Each of these levels could have sub-levels, for which conditional-probability models could be given. Ultimately, we are interested in the posterior distribution:

[process, parameters|data] ∝ [data|process, parameters] ×[process|parameters] ×[parameters]

Excerpted from Spatio-Temporal Statistics with R (Wikle et al.)

✴️ @AI_Python_EN

Everyone strives to build accurate and high-performing #datascience models. Check out these articles that list down the different ways to improve your model's accuracy and evaluate it:

8 Proven Ways for improving the “Accuracy” of a #MachineLearning Model - https://buff.ly/2TP4sGI

Improve Your Model Performance using Cross Validation (in Python and R) - https://buff.ly/2HGNV05

7 Important Model Evaluation Error Metrics Everyone should know - https://buff.ly/2HsYgxm

✴️ @AI_Python_EN