Full Stack Deep Learning covers the full lifecycle of an AI application, from ideation through deployment but it does not cover theory or model fitting.

There is a lot of hype around Artificial Intelligence (AI) and as I mentioned in my last blog there is weak and strong AI. Personally I prefer the term narrow instead of weak for AI as the AI we have today is focused on narrow or very specific tasks. AI in of itself is a large field and the best way of getting to understand AI is to get a grounding in the basics. I like to think of AI as being an onion, as it has layers of depth starting on the outside with AI and with Deep Learning at the centre. Don't worry I will explain what these terms mean in a little bit.

New AI-powered farming machines trained on the PyTorch framework are being developed to help farmers produce more food with fewer resources. Blue River Technology is using the PyTorch machine-learning framework to train robotic crop sprayers to identify and map weeds as they move through a field. Using a high-resolution camera array, the system instructs the machines exactly where to spray herbicide, killing weeds while leaving precious crops unharmed. Artificial intelligence in the real world: What can it actually do? What are the limits of AI?

This is a Stanford collected Dataset and is available for the public to experiment and to learn. SVHN is a real-world image dataset for developing machine learning and object recognition algorithms with minimal requirement on data preprocessing and formatting. It can be seen as similar in flavor to MNIST(e.g., the images are of small cropped digits), but incorporates an order of magnitude more labeled data (over 600,000 digit images) and comes from a significantly harder, unsolved, real world problem (recognizing digits and numbers in natural scene images). SVHN is obtained from house numbers in Google Street View images. The images are, in no way, preprocessed or ready to be used yet.

In 1950, Alan Turing developed the Turing Test as a test of a machine's ability to display human-like intelligent behavior. "Are there imaginable digital computers which would do well in the imitation game?" In most applications of AI, a model is created to imitate the judgment of humans and implement it at scale, be it autonomous vehicles, text summarization, image recognition, or product recommendation. By the nature of imitation, a computer is only able to replicate what humans have done, based on previous data. This doesn't leave room for genuine creativity, which relies on innovation, not imitation.

"GPT-3 is not a mind, but it is also not entirely a machine. It's something else: a statistically abstracted representation of the contents of millions of minds, as expressed in their writing." In recent years, the AI circus really has come to town and we've been treated to a veritable parade of technical aberrations seeking to dazzle us with their human-like intelligence. Many of these sideshows have been "embodied" AI, where the physical form usually functions as a cunning disguise for a clunky, pre-programmed bot. Like the world's first "AI anchor", launched by a Chinese TV network and -- how could we ever forget -- Sophia, Saudi Arabia's first robotic citizen.

Robots are extending their presence in domestic environments every day, being more common to see them carrying out tasks in home scenarios. In the future, robots are expected to increasingly perform more complex tasks and, therefore, be able to acquire experience from different sources as quickly as possible. A plausible approach to address this issue is interactive feedback, where a trainer advises a learner on which actions should be taken from specific states to speed up the learning process. Moreover, deep reinforcement learning has been recently widely utilized in robotics to learn the environment and acquire new skills autonomously. However, an open issue when using deep reinforcement learning is the excessive time needed to learn a task from raw input images. In this work, we propose a deep reinforcement learning approach with interactive feedback to learn a domestic task in a human-robot scenario. We compare three different learning methods using a simulated robotic arm for the task of organizing different objects; the proposed methods are (i) deep reinforcement learning (DeepRL); (ii) interactive deep reinforcement learning using a previously trained artificial agent as an advisor (agent-IDeepRL); and (iii) interactive deep reinforcement learning using a human advisor (human-IDeepRL). We demonstrate that interactive approaches provide advantages for the learning process. The obtained results show that a learner agent, using either agent-IDeepRL or human-IDeepRL, completes the given task earlier and has fewer mistakes compared to the autonomous DeepRL approach.

Deep reinforcement learning has shown remarkable success in the past few years. Highly complex sequential decision making problems have been solved in tasks such as game playing and robotics. Unfortunately, the sample complexity of most deep reinforcement learning methods is high, precluding their use in some important applications. Model-based reinforcement learning creates an explicit model of the environment dynamics to reduce the need for environment samples. Current deep learning methods use high-capacity networks to solve high-dimensional problems. Unfortunately, high-capacity models typically require many samples, negating the potential benefit of lower sample complexity in model-based methods. A challenge for deep model-based methods is therefore to achieve high predictive power while maintaining low sample complexity. In recent years, many model-based methods have been introduced to address this challenge. In this paper, we survey the contemporary model-based landscape. First we discuss definitions and relations to other fields. We propose a taxonomy based on three approaches: using explicit planning on given transitions, using explicit planning on learned transitions, and end-to-end learning of both planning and transitions. We use these approaches to organize a comprehensive overview of important recent developments such as latent models. We describe methods and benchmarks, and we suggest directions for future work for each of the approaches. Among promising research directions are curriculum learning, uncertainty modeling, and use of latent models for transfer learning.

Hex is a complex game with a high branching factor. For the first time Hex is being attempted to be solved without the use of game tree structures and associated methods of pruning. We also are abstaining from any heuristic information about Virtual Connections or Semi Virtual Connections which were previously used in all previous known computer versions of the game. The H-search algorithm which was the basis of finding such connections and had been used with success in previous Hex playing agents has been forgone. Instead what we use is reinforcement learning through self play and approximations through neural networks to by pass the problem of high branching factor and maintaining large tables for state-action evaluations. Our code is based primarily on NeuroHex. The inspiration is drawn from the recent success of AlphaGo Zero.

In many applications of machine learning (ML), updates are performed with the goal of enhancing model performance. However, current practices for updating models rely solely on isolated, aggregate performance analyses, overlooking important dependencies, expectations, and needs in real-world deployments. We consider how updates, intended to improve ML models, can introduce new errors that can significantly affect downstream systems and users. For example, updates in models used in cloud-based classification services, such as image recognition, can cause unexpected erroneous behavior in systems that make calls to the services. Prior work has shown the importance of "backward compatibility" for maintaining human trust. We study challenges with backward compatibility across different ML architectures and datasets, focusing on common settings including data shifts with structured noise and ML employed in inferential pipelines. Our results show that (i) compatibility issues arise even without data shift due to optimization stochasticity, (ii) training on large-scale noisy datasets often results in significant decreases in backward compatibility even when model accuracy increases, and (iii) distributions of incompatible points align with noise bias, motivating the need for compatibility aware de-noising and robustness methods.