Amir Ziai·Dec 8, 2024The algo innovation dilemmaExplore the “Algo Innovation Dilemma” — balancing algorithmic research with business impact. Discover strategies for sustainable success.
Amir Ziai·Dec 8, 2024Algo productization perception vs. realityReality of the breakdown of efforts for building algos, especially impactful ones, is very different than the common perception.
Amir Ziai·Aug 19, 2024mlbench — ML benchmarking utilsSimple utilities and visualizations to help with doing the right thing when you’re running ML experiments.
Amir Ziai·Feb 6, 2024An impact assessment framework for human-in-the-loop algo projectsThis post presents a framework for assessing the impact of human-in-the-loop project, in addition to adoption pitfalls and some…
Amir Ziai·Jan 3, 2024Load balancing for Data Parallel Actor-based systemsWe explore ways to dynamically load balance actors in a streaming framework.
Amir Ziai·Dec 30, 2023Retrieval-Augmented Few-shot Image Classification (RAFIC)A method for augmenting support images for few-shot image classification with additional retrieved images, with yields sig improvement.
Amir Ziai·Sep 20, 2023Ranking metrics from first principles: Average PrecisionHaven’t read part 1 yet? Ranking metrics from first principles lays down some of the required foundations that you may find useful when…
Amir Ziai·Sep 12, 2023A semi-random collection of advice that I’ve found very useful as an ML engineer1- Own things e2e 2- engg best practices 3- metrics! 4- problem formulation 5- bulletproof eval code 6- earn your complexity
Amir Ziai·May 16, 2023Reservoir samplingSample a large data stream with optimal storage and minimal compute.
Amir Ziai·Apr 12, 2023Count-min sketchNeed to count elements in an unbounded data stream using bounded memory? May want to consider count-min sketch.
