AI, Python, Cognitive Neuroscience

Urban legends about regression abound...Regression is much more than OLS linear or binary logistic regression.

Normally distributed data are not a requirement. Regression is not limited to modeling linear relationships. Data correlated across time or cross-sectionally can be accommodated, as can data from mixed distributions.

Some extensions of regression are able to simultaneously model multiple dependent variables of any data type - continuous, interval, ordinal, nominal and count. Some can include latent (unobserved) variables on either side of the equation.

For those with a basic background in regression, this book by two leading authorities, is an excellent next step.
https://www.crcpress.com/Generalized-Linear-Models-and-Extensions-Fourth-Edition/Hardin-Hilbe/p/book/9781597182256

✴️ @AI_Python_EN

--Activation Function in CNN--
-Image Analysis-

#imageanalysis #machinelearning #clustering #datascientists #kmeans #deeplearning #neuralnetwork #underfitting

We are Available in Persian

Python and Linux :

https://t.me/PythonLinuxExperts

DeepLearning and Artificial Intelligence:

https://t.me/DeepLearningAIExperts

Natural Language Processing and Data Mining:

https://t.me/NLPExperts

Busy Programmer Designs Chatbot To Talk To His Girlfriend While At Work
https://in.mashable.com/tech/4051/busy-programmer-designs-chatbot-to-talk-to-his-girlfriend-while-at-work

Busy boyfriend creates bot to chat with his girlfriend
https://www.abacusnews.com/digital-life/busy-boyfriend-creates-bot-chat-his-girlfriend/article/3013843

from: cvision

18 Impressive Applications of Generative Adversarial Networks (GANs)
A Generative Adversarial Network, or GAN, is a type of neural network architecture for generative modeling

There are two main types of consumer segmentation.

The first is unsupervised (post hoc) in which segments are "discovered" with some form of cluster analysis. K-means and hierarchical clustering historically have been most popular, though there are many newer methods such as latent class that are also used.

The second type is supervised (a priori) in which a target (dependent) variable is statistically profiled. The target variable is usually (though not always) a nominal or ordinal variable. Logistic regression, CHAID and random forests are three of many analytics tools used.

Often the two are used in combination, e.g., post hoc segmentation to discover the segments and a priori segmentation to predict them. In this case, the post hoc clusters are the target variable.

A third kind, which I call driver segmentation, essentially combines the two approaches simultaneously. This is less known in marketing research but can be very useful.

Brand segmentation is somewhat different. In brand segmentation, brand "maps" are created using correspondence analysis, biplots, principal components, hierarchical clustering or other methods to identify clusters of brands (not people).

✴️ @AI_Python_EN

Machine Learning for Artists
#DeepLearning #MachineLearning #ArtificialIntelligence #neuralnetwork #gan

http://ml4a.github.io/

✴️ @AI_Python_EN

Exercise safely with AI and Computer Vision

In partnership with University of Zurich, startup VAY has just created a new way to coach fitness, using Artificial Intelligence and Computer Vision. The app called "Vay Sport" helps to avoid injuries and improve performance while training. It observe the exercises and provides real-time feedback on posture during workouts

Thanks to deep learning, the App instantly creates a computer model of the human body to read out joint angles and limb positions.
The algorithm was developed with certified coaches to recognize optimal exercise execution

Read more here: https://lnkd.in/fM9WmmN

#deeplearning #computervision #artificialintelligence #training #fitness #innovation #technology

✴️ @AI_Python_EN

Nice project by Hugging Face (https://huggingface.co/) called "Write With Transformer" where you can write anything in a Google Doc-like interface and then get text completions (GPT-2 language model) multiple times. I've just tried it out and it's really cool. Check it out! #deeplearning #machinelearning

Link: https://lnkd.in/dGHqYDa

✴️ @AI_Python_EN

The Neural Network Visualization course has some solid momentum! 20 (out of 33) coding exercises have now been released. It's now US$4, but will be US$6 when it's finished, so you can still get it at a discount. Come preview the first section for free.

https://lnkd.in/eNeue6N

✴️ @AI_Python_EN

Consider checking out this Amazing tutorial series on AI and Machine learning on YouTube.

#machinelearning #artificialintelligence #deeplearning #python #computervision

https://lnkd.in/eui_KjZ

AI and Machine Learning Part 1 of 4

✴️ @AI_Python_EN

Many statistical procedures make the assumption that the data (observations) are independent and identically distributed (i.i.d.).

Often, however, this assumption is unrealistic and statisticians have developed numerous methods appropriate for situations when it is untenable.

For example, employee attitudes in one company are usually more similar within that company than to attitudes in another company. Hierarchical models (aka multilevel models) can be used to account for this lack of independence in the data.

Data may be correlated across time, too. For example, sales in one week are usually similar to sales in the preceding week. Many methods have been developed to account for autocorrelated data and are widely used in many disciplines.

Mixture models are useful when we cannot assume that one distribution applies to all observations. Some data are multimodal, for instance, and mixture models can account for this.

Note that we need not assume one distribution applies to all variables in our data. It is possible to model continuous, ordinal, nominal and count data simultaneously, for example. Hierarchical, time-series and mixture models can all be combined too, if necessary.

I have no idea how many ways there are to conduct a regression type of analysis. Hundreds would not be an exaggeration.

✴️ @AI_Python_EN

Remember that Mona Lisa animation? Now a new deep learning system can make an onscreen speaker say anything you type. Read the paper by Stanford, Princeton, and Adobe researchers:
http://bit.ly/2WIsftb
#DeepLearning #GANs #MachineLearning #ArtificialIntelligence

related post :
https://t.me/ai_python_en/1413
and
https://t.me/ai_python_en/1274
✴️ @AI_Python_EN

Statistics makes heavy use of the normal distribution for mathematical reasons, and also because the assumption of normality is often reasonable. Many natural phenomena are normally distributed, in fact.

That said, statisticians make use of dozens of distributions. As Bayesian statistics becomes mainstream, it's more important than ever that statisticians and data scientists have a good understanding of probability.

I've been cracking the books myself, and some books I've found helpful include:

- Introduction to Probability (Bertsekas and Tsitsiklis)
- Introduction to Probability Models (Ross)
- Essentials of Probability Theory for Statisticians (Proschan and Shaw)
- Handbook of Statistical Distributions (Krishnamoorthy)
- Statistical Distributions (Forbes et al.)

Hardin and Hilbe's book in the Generalized Linear Model is an excellent overview of GLMs. In addition to hierarchical modeling and mixture models, there are many extensions such as GAM, VGAM and GAMLSS, Quantile Regression, etc. Frank Harrell's book Regression Modeling Strategies is a gold mine, IMO.

There are also SEM extensions of the GLM, not to mention factor analysis, latent class clustering and a gigantic number of other unsupervised methods.

I can also recommend the Journal of the American Statistical Association to those who believe statistics is outdated and inflexible...:-) Lots of crazy urban legends circulating in some corners of the data science community.

✴️ @AI_Python_EN

DeepFake video concerns. AI researchers designing systems to detect fabricated footage of real people known as “deepfakes.” https://lnkd.in/gDFsGVq #machinelearning #deeplearning #deepfake #artificialintelligence… https://www.instagram.com/p/ByrUkWDnsgi/?igshid=1fum3iz1bmd5c …

✴️ @AI_Python_EN

At #CVPR2019, NVIDIA Researchers will present 20 accepted papers, including 11 orals. Check out the list here: https://nvda.ws/2WzDaAK

✴️ @AI_Python_EN

Wild idea from Google folks - taking compression/pruning/quantization to the next level by turning neural nets into simple look-up tables. "Table-Based Neural Units: Fully Quantizing Networks for Multiply-Free Inference," Covell et al.: https://arxiv.org/abs/1906.04798

✴️ @AI_Python_EN

"Write With Transformer" project, : text completions (GPT-2 language model) .
https://transformer.huggingface.co/ #deeplearning #machinelearning

Join me at TensorFlow + Swift + OpenAI's Unsupervised Sentiment, KubeFlow, TFX, Kubernetes http://meetu.ps/e/FXy9j/HydPD/a