Practical AI

2019 was a landmark year for AI, featuring major breakthroughs like BERT and GPT-2, which revolutionized natural language processing. The hosts dive into amazing feats, such as a robotic hand solving a Rubik's Cube with finesse. They also tackle the ethical dilemmas posed by advancements like deepfakes and emphasize the need for responsible AI practices. As they look towards 2020, discussions about the integration of diverse perspectives and the societal impacts of AI innovations take center stage, promising an exciting future.

Andrew Stanton, the product manager of search ranking at Etsy, shares his 15 years of experience in machine learning and search technology. He delves into how AI is transforming search in e-commerce, discussing the challenges of tailoring search for various handmade and vintage items. Andrew introduces neuroevolution, a method aimed at enhancing search algorithms, and emphasizes the benefits of using Rust programming for ML systems. He also shares insights on future trends in AI and its impact on search technology.

Evan Sparks, Co-founder and CEO of Determined AI, sheds light on the prevalent issues in AI infrastructure. He highlights why many are still in the 'dark ages' and discusses innovative solutions like fault-tolerant training and AutoML. Evan shares his journey from finance to AI, emphasizing the transition challenges. He also addresses the fragmentation in AI workflows and the need for cohesive systems that improve productivity and reproducibility. With an optimistic view, he explores the positive impacts of AI on the economy and environmental health.

Ines Montani and Matthew Honnibal, core developers of the powerful NLP library spaCy and co-founders of Explosion AI, dive into the fascinating world of natural language processing. They share SpaCy's journey, highlighting its adoption and unique features. The duo discusses the critical role of data annotation and efficient tooling in machine learning, while demystifying complex NLP models using a four-step understanding framework. They also touch on practical workflows, ethical considerations, and the vibrant community surrounding NLP.

Jakub Langr and Vladimir Bok, authors of "GANs in Action" and experts on Generative Adversarial Networks, dive into the transformative world of GANs. They unpack the basics and unique training processes that set GANs apart from traditional neural networks. The discussion highlights their revolutionary applications in creating photorealistic imagery and explores innovative types like CycleGAN and StyleGAN. Additionally, they address the ethical implications of GAN technology, including the risks of deepfakes, while discussing practical entry points for newcomers.

Adrien Treuille, Co-founder and CEO of Streamlit, discusses the revolutionary app framework designed for machine learning and data science. He highlights how Streamlit simplifies the development of custom ML applications, allowing non-technical users to create interactive UIs effortlessly. Adrien shares insights on its rapid adoption by major companies and contrasts Streamlit with traditional tools, emphasizing its versatility and community-driven innovations. He also touches on the ability to transform complex data scripts into user-friendly interfaces.

James Fletcher, Principal Scientist at Grakn Labs and an expert in machine intelligence, dives into the fascinating world of knowledge graphs. He clarifies how knowledge graphs differ from traditional databases and provides practical tips on building your own. The conversation also explores the synergy between AI and knowledge graphs, including enhancing tasks like computer vision. James discusses advanced techniques, such as graph neural networks, and showcases their potential in fields like drug discovery, emphasizing the future of intelligent systems.

A breakthrough in AI allows robotic hands to tackle the Rubik's Cube with impressive skill. The conversation dives into the challenges of training robots for precise tasks and the clever algorithms behind their success. With a mix of humor and technical insights, the hosts critique the hype surrounding AI developments and the public’s perceptions. They also discuss the vital role of simulation in robotics, blending education with engaging anecdotes. Listeners are encouraged to explore further learning resources on AI and machine learning.

Michael Malyuk, CEO of Heartex and a contributor to Label Studio, shares his expertise on the crucial yet challenging process of data labeling in AI. He discusses the significant impact of accurate labeling on model performance and highlights various open-source tools that aim to streamline the process. Malyuk addresses common pitfalls, including time consumption and bias, and explores innovative solutions like Label Studio that enhance collaboration and customization for users. The conversation sheds light on the evolving landscape of data labeling and its critical role in AI development.

Anais Dotis-Georgiou, an expert at InfluxData, dives into the fascinating world of time series data. She explains its omnipresence, from stock prices to server logs and even photos on our phones. Anais shares valuable insights on the effectiveness of traditional statistical methods versus neural networks for time series analysis. The conversation also highlights practical applications, including anomaly detection and data querying techniques using InfluxDB, making complex topics accessible for data enthusiasts.