The Receptive Field as a Regularizer in Deep Convolutional Neural Networks for Acoustic Scene Classification. Khaled Koutini, Hamid Eghbal-zadeh, Matthias Dorfer, and Gerhard Widmer
http://arxiv.org/abs/1907.01803

✴️ @AI_Python_EN

Data at its lowest form of aggregation is not always most useful, and it's very easy to waste time, energy and computing resources trying to model and predict tiny patterns and fluctuations which are, for all intents and purposes, random.

Some big data are unnecessary big, and more value at less cost could have been extracted had the data been intelligently aggregated prior to the modeling.

There is also the habit of using gigantic data files when samples would have sufficed. :-)

✴️ @AI_Python_EN

If you're interested in transfer learning in NLP, you definitely need to check out Daniel Pressel's Github repository. He recently gave a keynote and tutorial at the International Summer School on Deep Learning, 2019 in Gdansk, Poland. The slides are online and really amazing and also he uploaded the tutorial. Colab notebooks are also provided. Check it out! #deeplearning #machinelearning

📊 Slides: https://lnkd.in/dttQJ8P
🔤 Github: https://lnkd.in/d_fscpV

✴️ @AI_Python_EN

https://lnkd.in/fJkMyHM

Facebook AI Research (FAIR) and NYU School of Medicine’s [Center for Advanced Imaging Innovation and Research (CAI²R)](http://www.cai2r.net/) are sharing new open source tools and data as part of [fastMRI, a joint research project to spur development of AI systems to speed MRI scans by up to 10x](https://lnkd.in/fQxPgMv) (3 minutes instead of 30 minutes).
✴️ @AI_Python_EN

Learning from unlabelled videos is key to scaling computer vision models. Our method learns state-of-the-art pose detectors given only examples of skeletons or facial landmarks. It further enables image generation and manipulation conditioned on landmarks.
https://arxiv.org/abs/1907.02055

✴️ @AI_Python_EN

Benchmarking Model-Based Reinforcement Learning"
https://arxiv.org/pdf/1907.02057.pdf
summarized existing MBRL algorithms into three types, and benchmarked 14 algorithms on 18 environments (CartPole to Humanoid).
Project & Code http://www.cs.toronto.edu/~tingwuwang/mbrl.html

✴️ @AI_Python_EN

bit.ly/2JeYsQr
Demystifying the Math Behind Neural Nets — learn how #NeuralNetworks Learn, with an implementation demonstrated one step at a time:
http://bit.ly/2JgmGKh

✴️ @AI_Python_EN

DeepMind:
For anyone interested in constrained optimisation with DL models (e.g. as in https://arxiv.org/abs/1810.00597 ), we just released a few handy tools to deal with inequality constraints for Sonnet (http://tiny.cc/a9va9y ).

✴️ @AI_Python_EN

New paper on much faster RL by learning graph representations of the world http://arxiv.org/abs/1907.00664 ! World graphs capture the structure of the world and can be used to focus exploration:
http://www-personal.umich.edu/~shangw/world_graph.html

✴️ @AI_Python_EN

Training on only synthetic data, our new paper gets near the SOTA on the real-world 3DPW human pose dataset by focusing on motion. Simulation is underappreciated for representation learning; you just have to keep the nets from overfitting to the domain.
https://arxiv.org/abs/1907.02499

✴️ @AI_Python_EN

Current autopilot plane landings are not really autonomous: they rely on Instrument Landing Systems (ILS), only available at major airport runways. Computer Vision enables true autonomous landings: finding & landing on runway without ground instruments.
https://www.tum.de/nc/en/about-tum/news/press-releases/details/35556/

✴️ @AI_Python_EN

excited to announce the release of #Trafffic-Net, an image dataset for training real-time traffic analysis and accident detection #ArtificialIntelligence systems. https://github.com/OlafenwaMoses/Traffic-Net

✴️ @AI_Python_EN

#DeepLearning: What Is It Good For? - Prof. Ankit Patel Rice University [Video]
https://buff.ly/2U1irV9
#AI #MachineLearning

✴️ @AI_Python_EN

Psychometricians make use of a very wide range of statistical models, for example:

Regression analysis
Path analysis
Exploratory factor analysis
Confirmatory factor analysis
Item response theory modeling
Structural equation modeling
Growth modeling
Discrete-time survival analysis
Continuous-time survival analysis
Time series analysis

Observed outcome variables may be continuous, censored, binary, ordered categorical (ordinal), unordered categorical (nominal), counts, or combinations of these variable types.

Some variations of these basic models are:

Regression mixture modeling
Path analysis mixture modeling
Latent class analysis
Latent class analysis with covariates and direct effects
Confirmatory latent class analysis
Latent class analysis with multiple categorical latent variables
Loglinear modeling
Multiple group analysis
Multilevel modeling
Finite mixture modeling
Complier Average Causal Effect (CACE) modeling
Latent transition analysis and hidden Markov modeling including mixtures and covariates
Latent class growth analysis
Discrete-time survival mixture analysis
Continuous-time survival mixture analysis

Over the years, I've used many of these models in my marketing-related work. They all have practical applications in many fields - it ain't just academic stuff.

✴️ @AI_Python_EN

A PyTorch implementation of "SimGNN: A Neural Network Approach to Fast Graph Similarity Computation" (WSDM 2019). This is a lightweight graph convolutional neural network for the fast calculation of approximate graph similarity at scale. Graph similarity search is among the most important graph-based applications, e.g. finding the chemical compounds that are most similar to a query compound.

https://lnkd.in/gA5tfuC

#datamining #machinelearning #deeplearning #datascience #bigdata

✴️ @AI_Python_EN

Great Statistical software for Beginners.

Here is the Gretl Tutorial by Simone Gasperin

1)Simple Linear Regression
https://lnkd.in/ecfsV9c

2)Coding Dummy Variables
https://lnkd.in/ef7Yd7f

3)Forecasting New Observations
https://lnkd.in/eNKbxbU

4)Forecasting a Large Number of Observations
https://lnkd.in/eHmibGs

5)Logistic Regression
https://lnkd.in/eRfhQ87

6)Forecasting and Confusion Matrix
https://lnkd.in/eaqrFJr

7)Modeling and Forecasting Time Series Data
https://lnkd.in/e6fqKpF

8)Comparing Time Series Trend Models
https://lnkd.in/eKjEUAE

#datascience #machinelearning #statistics #dataanalytics #dataanalysis

✴️ @AI_Python_EN

What metrics to focus on, in confusionmatrix?
Ans-> It depends on the problem statement and data one is dealing with!

#examples
1) Spam Filter -> Consider +ve class as 'spam'. Optimize for Precision/Specificity, the reason for the same is...
False Negatives(spam emails in the primary) are more acceptable than False Positives(primary emails in Spam).

2) Fraud Transactions -> Consider +ve class as 'Fraud'. Optimize for Sensitivity, the reason for the same is...
False Positives(detecting fraud, but they are not) are more acceptable than False Negatives(detecting not as a fraud, but actually they are).

Will discuss more of Classification problem in our group

#datascience #machinelearning

✴️ @AI_Python_EN

https://youtu.be/9M18rc9-VWU

https://github.com/yuanming-hu/taichi_mpm

✴️ @AI_Python_EN

Implementation of Variational Auto-Encoder (VAE) and Deep Feature Consistent VAE for facial attribute manipulation using Keras and Tensorflow-dataset module.

https://github.com/iamsoroush/face_vae

✴️ @AI_Python_EN

A Full Hardware Guide to Deep Learning

🌎 http://bit.ly/2CBNi3W

#DeepLearning #MachineLearning #AI #DataScience

✴️ @AI_Python_EN

Is this AI developing a sense of time?

https://www.zdnet.com/article/is-this-ai-developing-a-sense-of-time/

✴️ @AI_Python_EN