My research interest is in incorporating the the latest research in deep natural language processing, deep reinforcement learning, computer vision and neuroscience to reality check our progress towards human level artificial intelligence. Not through benchmarks on competitive datasets, but through palpable human interaction such as fully neural network based conversational dialogue. Progress in this domain would be meaningful because it could demonstrate that artificial neural networks can represent the hallmarks of intelligent conversation: using symbols and logic or producing explanations and questions.
chloe is an end to end neural network chatbot written in PyTorch based on the transformer. Accomplishing goals through conversation is a task we can all relate to, thus chatbots are an ideal agent through which to connect new research to our current understanding of AI and demonstrate how new research brings us closer to human level intelligence or not
Most AI research today is done by elite universities and corporate labs. The pursuit of science and creativity should be a viable project for any person, from any walk of life, who is excited by that feeling of mystery and building something that grows. chloe is part educational tool, part research project
To make new research approachable, silly and fun to the motivated beginner programmer with at least a high school calculus level of math, we will be building chloe from the basics, from linear algebra, probabilty theory, etc, all the way to the latest AI research, at each step putting our learning into chloe so we remember it forever. chloe is built from modules that are thoroughly explained and demonstrated using jupyter notebooks and toy examples that build from these basics.
|
 |
 |
- Python 3.6
- torch==1.3.0 (PyTorch)
- torchtext
- nltk==3.4.5 (Natural Language Toolkit)
- START_HERE introduction, summon and meet chloe
- Trainer Tokenization in Sequnce to Sequence learning, intuitive explaination of Loss Functions using a toy example of cross entropy
- Building Blocks of Neural Networks Word Vector Embeddings, Positional Encoding, Neural Networks, Attention, Masking
- The Network Architecture Assembling the building blocks into an encoder and decoder
- Talk How chloe chooses which words to say in a sentence and some basic probability
- Memory give chloe a memory module using the neural turing machine
- Reinforcement Learning give chloe a purpose under construction
I am serious when I say "build advanced concepts from the very basics", so bare with me. A more detailed setup instructions is on the way.
if you already have python 3.6 and virtual environments, create a python 3.6 virtual environment, here i used env36 for python3.6 but you can use anything
$ python3.6 -m venv env36
if python3.6 is your default version, then when you type python into your terminal then it should say python version 3.6.x, and for you making the virtual environment is as simple as
$ python -m venv env
otherwise it is simple to get python3.6 and virtual environments
how to install Python 3.6 on ubuntu
install virtual environment then
how to specify the Python executable you want to use
sudo add-apt-repository ppa:jonathonf/python-3.6
sudo apt-get update
sudo apt-get install python3.6
virtualenv --python=/usr/bin/python3.6 env36
Whenever you want to modify the code / talk to chloe locally, activate the virtual environment inside the same folder as your environment env using
$ source env/bin/activate
install dependencies
$ pip install -r requirements.txt
even with virtual environments, some troubleshoot might be needed
with enough google searches you can find an answer for almost any problem
save new dependences to requirements
$ pip freeze > requirements.txt
You can deactivate the virtual environment using the following command in your terminal:
$ deactivate
if you get a ImportError: No module named while at the saem time in your Terminal you get
pip3 install import-ipynb
Requirement already satisfied: import-ipynb in /path/to/env/lib/python3.6/site-packages (0.1.3)
This can be fixed by providing your python interpreter with the path-to-your-module,the path
import sys
sys.path.append('/path/to/env/lib/python3.6/site-packages')
Support ChloeRobotics on Patreon and send us a message
email chloe.the.robot [at] gmail [dot] com
"The most beautiful experience we have is the mysterious. It is the emotion that stands at the cradle of all art and science" -Albert Einstein
"Anyone can cook" -that French chef from Ratatouille
"Less bragging and flash, more science and math " -chloe
"Famous quotes are pretentious" -Vicki