More posts around deep learning and google deep mind

 robotics

http://hutter1.net/ai/suaibook.pdf

https://arxiv.org/abs/1708.04782

https://github.com/google-deepmind/pysc2/blob/master/pysc2/lib/named_array.py

https://github.com/google-deepmind/deepmind-research/tree/master/tandem_dqn

https://www.youtube.com/watch?v=hCeJeq8U0lo&t=9s

https://arxiv.org/pdf/2105.14039

https://arxiv.org/pdf/2102.05182

https://openreview.net/forum?id=nPHA8fGicZk

https://github.com/google-deepmind/deepmind-research/tree/master/tandem_dqn


. Neuromorphic Computing

Technologies: Neuromorphic chips like Intel's Loihi or IBM's TrueNorth are designed to mimic the structure and function of the human brain, making them ideal for brain-like AI models.

Purpose: These chips are designed to support spiking neural networks (SNNs), which are more biologically plausible than traditional artificial neural networks.

2. Spiking Neural Networks (SNNs)

Algorithms: SNNs use spikes (events) to transmit information, similar to how neurons communicate in the brain. This allows for more efficient and brain-like computation.

Purpose: SNNs are particularly useful for continuous learning and energy-efficient computation.

3. Reinforcement Learning (RL)

Algorithms: Algorithms like Deep Q-Networks (DQN), Proximal Policy Optimization (PPO), and Actor-Critic methods can be adapted for continuous learning environments.

Purpose: RL algorithms enable the model to learn from interaction with the environment over time, improving decision-making through trial and error.

4. Continual Learning (Lifelong Learning)

Technologies/Algorithms: Elastic Weight Consolidation (EWC), Synaptic Intelligence (SI), and Memory Aware Synapses (MAS) help models retain knowledge over time without forgetting previous knowledge (addressing the problem of catastrophic forgetting).

Purpose: These techniques are crucial for building a model that can learn continuously over years.

5. Self-supervised Learning

Algorithms: Techniques like Contrastive Predictive Coding (CPC) and SimCLR allow models to learn from unlabelled data by predicting parts of the data based on other parts.

Purpose: This is akin to how humans learn from raw sensory input without explicit labels, making it more aligned with how the brain learns.

6. Biologically Inspired Algorithms

Examples: Hebbian learning, inspired by the way synaptic connections strengthen based on experience, and Bayesian learning, which mimics the brain's probabilistic reasoning capabilities.

Purpose: These algorithms aim to replicate specific brain functions and learning processes.

7. Large-scale Data Ingestion and Processing

Technologies: Distributed computing frameworks like Apache Spark, Kubernetes, and cloud platforms (e.g., AWS, Google Cloud) for handling and processing massive amounts of data over time.

Purpose: To support the long-term and continuous absorption of data at scale.

8. Neuroscience-Inspired Architectures

Examples: Models inspired by cortical columns, the hippocampus, and other brain structures can help build more biologically plausible architectures.

Purpose: These architectures are designed to replicate specific brain functions, such as memory formation and spatial reasoning.

9. Hardware Acceleration

Technologies: GPUs, TPUs, and specialized AI accelerators like Google's Tensor Processing Units (TPUs) or FPGA-based systems.

Purpose: To support the computational demands of continuous learning over extended periods.

10. Ethics and Safety Considerations

Frameworks: Implementing ethical AI frameworks and ensuring safe exploration in RL to prevent harmful behaviors.

Purpose: Continuous learning AI systems need to be safe and aligned with ethical standards to prevent undesirable outcomes over long periods.



=====

https://www.reddit.com/r/MachineLearning/comments/1mhkx4/a_look_at_nupic_a_selflearning_ai_engine_part/


https://www.youtube.com/watch?v=VyMhZhwe3gc



https://github.com/htm-community/NAB



https://github.com/orgs/htm-community/repositories


https://stanfordmag.org/contents/keeping-secrets



https://www.numenta.com/blog/2023/07/28/top-5-questions-from-customers/




======

https://www.maragu.dev/blog/go-is-my-hammer-and-everything-is-a-nail



https://www.jefftk.com/p/history-of-https-usage


https://htmchallenge.devpost.com/details/resources



/////


https://devpost.com/software/htm-models-adelaide




https://github.com/iizukak/ecg-htm




https://devpost.com/software/ecg-htm


Comments

Post a Comment

Popular posts from this blog

Testing multi machine hadoop cluster

 So I was testing setup of a hadoop cluster and running map reduce. There are preliminary notes in first post. One key, I ran all commands under openssh as that user Setup the user Main machine: sudo usermod --shell /bin/bash mainhdfs Start openssh   openssh   su - mainhdfs   sudo apt-get install openssh-server   sudo systemctl enable ssh   sudo systemctl enable ssh --now   sudo systemctl start ssh   su - mainhdfs   sudo mkdir /usr/local/hadoop   sudo mv hadoop-3.3.6 /usr/local/hadoop/   sudo chown own -R mainhdfs:hadoop /usr/local/hadoop   sudo chown -R mainhdfs:hadoop /usr/local/hadoop Setup more on ssh: ssh-keygen -t rsa -P '' -f ~/.ssh/id_rsa cat ~/.ssh/id_rsa.pub >> ~/.ssh/authorized_keys chmod 600 ~/.ssh/authorized_keys chmod 700 ~/.ssh Write the following .bashrc and .bash_profile for that user export HADOOP_HOME=/usr/local/hadoop/hadoop-3.3.6 export PATH=$PATH:$HADOOP_HOME/bin:$HADOOP_HOME/sbin export HADOOP_CONF_...
Read more

Random thoughts on AI

 Random AI Thoughts Posts - things I am looking at: scale ai Pinecone Perplexity Notion Hebbia Character AI --- web server web client stand alone -    limited html and http - validates your html and shit   very strict     -- tomcat based      -- my own ML-From-Scratch Description: This repository provides implementations of various machine learning algorithms from scratch in Python, without using libraries like Scikit-learn. It's educational and shows how algorithms like k-means clustering, decision trees, and linear regression work under the hood. Repository: ML-From-Scratch on GitHub You-Get Description: A simple and powerful tool for downloading media files from the web. It uses machine learning for analyzing and categorizing the downloaded media. Features: Supports various media sites, custom download options, and automated media organization. 3. OpenAI Gym Description: A toolkit for developing and comparing reinforcement learning algorit...
Read more

Getting Numenta htm.java to run

Image
 OK, that didn't take long, the fork ran for htm.java numenta nupic So if you want to run with latest Java, Eclipse have it build, run the unit tests and run an integration, try the fork here: mvn compile exec:java https://github.com/berlinbrown/htm.java/blob/master/integration/testing/reports/BASIC-INTEGRATION-RUN.txt
Read more