DoorDash’s Post

One of the simplest techniques for improving sensitivity in your experiments is eliminating or reducing dilution. Dilution is akin to throwing more hay in the haystack, thus making your ability to find a signal extremely challenging. In this blogpost, David Press and I explain dilution and provide some practical guidance on how to handle it. We also provide some examples showcasing how enormous is the impact from eliminating dilution. For ML applications, having a way to handle dilution can lead to 100x+ sensitivity gain.

One of the simplest techniques for improving sensitivity in your experiments is eliminating or reducing dilution. Dilution is akin to throwing more hay in the haystack, thus making your ability to find a signal extremely challenging. https://lnkd.in/dYAHEe5k

A simpler implementation of Stanford/Google's seminal paper "Generative Agents: Interactive Simulacra of Human Behavior". A good starting point to experiment LLM + memory stream (recency, importance, relevancy, and reflections). https://lnkd.in/gpQXruNX

Excited to announce our latest blog post on multi-goal path finding algorithm! 🚀 In this study, we introduce PKE-RRT: an efficient approach powered by a multi-task learning model to tackle the challenges of finding closed and collision-free paths in a certain order. 🌐🏞️ One of the key challenges in solving the multi-goal path finding problem is the lack of prior knowledge of local paths between vertices. However, our innovative PKE model is designed to provide accurate estimates of local path length, enabling us to determine the weights of the graph. 📊⚖️ To enhance the path-finding process, our approach also predicts a promising region and guideline as heuristics. This facilitates efficient exploration of the tree structure, allowing PKE-RRT to rapidly provide a sub-optimal solution. 🌳📋 Extensive numerical experiments have shown the exceptional performance of PKE-RRT when dealing with different numbers of goals. It outshines in terms of calculation time, path cost, sample number, and success rate! 💯 Curious to learn more about this groundbreaking algorithm? Check out the blog post here: [Link](https://bit.ly/3s86vrc) 📖 #pathfinding #algorithm #machinelearning #research

Part 4/4: 𝐄𝐱𝐩𝐞𝐫𝐢𝐦𝐞𝐧𝐭𝐚𝐥 𝐃𝐞𝐬𝐢𝐠𝐧: 𝐙𝐞𝐫𝐨𝐢𝐧𝐠 𝐢𝐧 𝐨𝐧 𝐲𝐨𝐮𝐫 𝐁𝐚𝐬𝐞𝐥𝐢𝐧𝐞 𝐚𝐧𝐝 𝐄𝐱𝐩𝐞𝐜𝐭𝐞𝐝 𝐈𝐦𝐩𝐚𝐜𝐭 In this final blog of the experimentation series, we will dive deep into the risks behind the decisions made during experimentation. We will understand the minimal defectable effect (MDE), and the risks behind MDEs in experimental design. Link: https://lnkd.in/gVHsxRPx If you’d like to begin by reading the other parts of the blog written by Manisha Arora and Julian Hsu, find the link in the comments! Follow PrepVector for more blogs. #data #experimentation

The unidentifiable code pattern that drives performance engineers crazy. Especially when an important release is around the corner, performance is dropping and logs make no sense. One of my favorite discoveries of the Product Science machine learning algorithm that we can now surface in a matter of a click. See our latest webinar for more details

Great article on applying prompt tuning by Valentin Leonhard Buchner - https://lnkd.in/dWUcdJJD. I was lucky enough to get a personal walk through of this research. It was fascinating to learn how small models are being trained to create the input to LLMs.

A nice startup course to get the basics & understand the foundations of Prompt Engineering. Also nice list of articles & read up material ! Just finished the course “Prompt Engineering: How to Talk to the AIs” by Xavier Amatriain! Check it out: https://lnkd.in/gVDyRggw #generativeai #largelanguagemodels #promptengineering.

Keep Track of Your Backtests with DVC’s Experiment Tracking - Part 4 of the tutorial on how to use DVC for experiment tracking, this time, with time series forecasting by Eryk Lewinson