Check out Off-Policy Classification, a new method to evaluate the performance of #reinforcementlearning agents trained entirely on data from prior agents, enabling selective testing of only the most promising models on real-world robots. Learn more below!

https://ai.googleblog.com/2019/06/off-policy-classification-new.html

✴️ @AI_Python_EN

Waymo just announced the release of large open dataset at #CVPR2019

https://waymo.com/open

✴️ @AI_Python_EN

Artificial Intelligence Based Cloud Distributor (AI-CD): Probing Low Cloud Distribution with a Conditional Generative Adversarial Network

https://arxiv.org/abs/1905.08700

✴️ @AI_Python_EN

The 2019 IEEE Conference on Computer Vision and Pattern Recognition

Best Papers

https://syncedreview.com/2019/06/18/cvpr-2019-attracts-9k-attendees-best-papers-announced-imagenet-honoured-10-years-later/

✴️ @AI_Python_EN

NLP + Deep Leaning checked. Was painfully awesome. Now what's next? Can't waste it... or maybe CNN and RL? #cs224n #deeplearning #NLP

🐼🤹‍♂️ pandas trick:
Two easy ways to reduce DataFrame memory usage:
1. Only read in columns you need
2. Use 'category' data type with categorical data

Example:
df = pd.read_csv('file.csv', usecols=['A', 'C', 'D'], dtype={'D':'category'})

#Python #DataScience

✴️ @AI_Python_EN

Predicting demands using AI? Just focus on the things that matter

✴️ @AI_Python_EN

Statistical methods can and are applied to qualitative data (e.g. text). The raw data first need to be converted to a form, or represented in a way, that can be quantitatively analyzed.

It can be done by man or, increasingly, machine. This would be the case whether cluster analysis, factor analysis, deep learning or some other method is used to analyze the data.

A very simple example is cluster or factor analyzing open end codes from a consumer survey, or using them in key driver regression.

This predates AI by many years, though machine coding would now be the preferred approach in many instances.

Sparse data can be a concern, but with rapid declines in the cost of survey data collection, this is now more feasible since very large samples can be collected at reasonable cost and within a realistic time frame.

To be clear, my original motivation for this post was not to push an alternative to standard close-ended questioning, but a response to confusion about AI I frequently encounter. However, with data of sufficient quality, complex analytics which tie attitudes, behavior and demographics together, perhaps combined simultaneously with segmentation, is possible with this alternative method. AI in some form may play a part but is not absolutely essential. The basic idea goes back nearly a century.

✴️ @AI_Python_EN

We Can All Become Video Game Characters With This AI

Video: https://www.youtube.com/watch?v=Y73iUAh56iI

Paper: https://arxiv.org/abs/1904.08379

I've been thinking a bit about the growing practice of fine-tuning generic pretrained models: first in computer vision, now NLP (highly recommend Sebastian Ruders great article on this http://ruder.io/nlp-imagenet/ )...Last time I mentioned this, people were skeptical that RL would be next.
✴️ @AI_Python_EN

✴️ @AI_Python_EN

Death by algorithm: the age of killer robots is closer than you think
🔗 https://www.vox.com/2019/6/21/18691459/killer-robots-lethal-autonomous-weapons-ai-war
#machinelearning

✴️ @AI_Python_EN

Machine Learning Zero to Hero
☞ http://bit.ly/2QCtsMj
#ai #tensorflow

✴️ @AI_Python_EN

All materials of berkeley ai Deep Unsupervised Learning now up:
https://sites.google.com/view/berkeley-cs294-158-sp19/home

✴️ @AI_Python_EN

Using #AI To Analyze Video As Imagery: The Impact Of Sampling Rate
https://buff.ly/2IBFRO8
#ArtificialIntelligence #MachineLearning #DeepLearning #robotics

✴️ @AI_Python_EN

Bayesian Optimization with Binary Auxiliary Information
https://deepai.org/publication/bayesian-optimization-with-binary-auxiliary-information … by Yehong Zhang et al.
#ReinforcementLearning #Hyperparameter

✴️ @AI_Python_EN

Best Paper Finalist (top 1% of accepted papers) Check it out!
http://openaccess.thecvf.com/content_CVPR_2019/html/Ribera_Locating_Objects_Without_Bounding_Boxes_CVPR_2019_paper.html

Comparison of different #MachineLearning approaches for neuroimaging data
Main take-aways - prediction accuracy increased once N ≥ 400
- Substantial effect of pipeline on accuracies: Is this the new p-hacking?
https://buff.ly/2NdaJcv

✴️ @AI_Python_EN

Theory of the Frequency Principle for General Deep Neural Networks.
http://arxiv.org/abs/1906.09235
✴️ @AI_Python_EN