🤖

Machine Learning Engineer

Newsletters sent once a week, unsubscribe anytime.

Published 8th July 2025

🔧 Company Engineering Blogs

Is Your Model Fairly Certain? Uncertainty-Aware Fairness Evaluation for LLMs (machinelearning​.apple​.com). Proposes UCerF metric for assessing fairness in LLMs by addressing model uncertainty; introduces new dataset for gender-occupation fairness evaluation

Making group conversations more accessible with sound localization (research​.google). SpeechCompass uses multi-microphone localization to improve mobile captioning with speaker diarization and directional guidance, enhancing group conversation accessibility

📚 Academic Research

Revisiting Learning Rate Control (arxiv:cs). Analyzes learning rate control in deep learning, comparing optimization methods and their reliability, emphasizing algorithm selection and new approaches like meta-learning

The Gauss-Markov Adjunction: Categorical Semantics of Residuals in Supervised Learning (arxiv:stat). Category theory reformulates machine learning, focusing on the Gauss-Markov Adjunction to clarify residuals and parameters in supervised learning models like linear regression

Parsimonious Gaussian mixture models with piecewise-constant eigenvalue profiles (arxiv:stat). Introducing parsimonious Gaussian mixture models with piecewise-constant covariance eigenvalue profiles, enhancing flexibility in statistical learning for unsupervised problems

Tensor Decomposition Networks for Fast Machine Learning Interatomic Potential Computations (arxiv:cs). Tensor decomposition networks replace Clebsch-Gordan tensor products with low-rank decompositions, improving computation speed for interatomic potential calculations

Targeted tuning of random forests for quantile estimation and prediction intervals (arxiv:stat). Novel tuning procedure for random forests enhances quantile estimation accuracy and prediction interval validity, focusing on minimizing quantile coverage loss

System-performance and cost modeling of Large Language Model training and inference (arxiv:cs). Performance-cost modeling for LLM training and inference, integrating compute techniques, memory optimizations, communication strategies, and topology-aware algorithms

Scaling Collapse Reveals Universal Dynamics in Compute-Optimally Trained Neural Networks (arxiv:cs). Universal loss curve behavior in compute-optimally trained neural networks observed through supercollapse, analyzed with scaling laws and SGD noise dynamics

Gold after Randomized Sand: Model-X Split Knockoffs for Controlled Transformation Selection (arxiv:stat). Model-X Split Knockoffs enhance FDR control for transformation selection in random designs, with applications in signal processing and Alzheimer's disease imaging

DeltaSHAP: Explaining Prediction Evolutions in Online Patient Monitoring with Shapley Values (arxiv:cs). DeltaSHAP: an XAI algorithm for online patient monitoring, improving explanation quality and computational efficiency using Shapley values for prediction evolution

Semi-supervised learning for linear extremile regression (arxiv:stat). Introduction of semi-supervised linear extremile regression to improve estimation efficiency in high-dimensional settings, addressing challenges in nonparametric methods

Towards a Signal Detection Based Measure for Assessing Information Quality of Explainable Recommender Systems (arxiv:cs). Objective metric for assessing Veracity in explainable recommender systems, focusing on Fidelity and Attunement using signal detection theory

MvHo-IB: Multi-View Higher-Order Information Bottleneck for Brain Disorder Diagnosis (arxiv:cs). MvHo-IB framework enhances fMRI diagnosis by modeling pairwise and higher-order interactions, utilizing O-information and a novel Brain3DCNN encoder

🔬 Research & Applications

Arc Virtual Cell Challenge: A Primer (fleetwood​.dev). Explore the Arc Virtual Cell Challenge: gene silencing prediction using CRISPR, neural networks, RNA sequencing, and innovative modelling approaches from a non-biologist's perspective

Augmented Commerce: Machine Learning at Shopify (shopifyengineering​.myshopify​.com). Shopify employs machine learning tools like Qwen models and LLaMa to enhance merchant success, focusing on product classification, fraud assessment, and financial forecasting

AI system helps researchers unlock hidden potential in newly discovered materials (news​.engineering​.utoronto​.ca). U of T Engineering's AI tool predicts applications for newly developed metal-organic frameworks (MOFs), enhancing material discovery for carbon capture and drug delivery

Huawei releases an open weight model trained on Huawei Ascend GPUs (arxiv​.org). Huawei's Pangu Pro MoE introduces a Mixture of Grouped Experts leveraging efficiency and sparsity, trained on Ascend GPUs for advanced computational language tasks

👁️ Computer Vision & Graphics

Introduction to 3D Gaussian Splatting and Its Speedup (blog​.us​.fixstars​.com). 3D Gaussian Splatting optimized for rendering speed using Mahalanobis distance, resulting in a 15% speedup without quality loss in novel-view synthesis applications

AI-Enhanced Social Engineering Detection Defending Against Deepfake Attacks (blackmoreops​.com). Implement AI-enhanced detection for deepfake attacks using machine learning, audio and video analysis, behavioural patterns, and integration with security systems like SIEM

Interactive Data Exploration for Computer Vision Projects with Rerun (towardsdatascience​.com). Interactive data exploration for computer vision using Rerun, OpenCV, and Python enhances algorithm decision-making and visualizes complex signals efficiently

Overthinking GIS (2024) (scottsexton​.co). Exploring GIS usability through grade calculation, using DEM source data, vector calculus, and OpenCV to analyze terrain steepness and create usability maps

🖥️ Hardware & Infrastructure

Seventh-generation server hardware at Dropbox: our most efficient and capable architecture yet (dropbox​.tech). Dropbox's seventh-generation server hardware features Crush, Dexter, Sonic, Gumby, and Godzilla, enhancing performance, storage density, and AI capabilities

How AI on Microcontrollers Actually Works: The Computation Graph (danielmangum​.com). Explores TensorFlow Lite for microcontrollers, computation graphs, model file formats, and inference performance with practical examples using YOLO models

Navigating Failures in Pods With Devices (kubernetes​.io). Insights on handling failures in Kubernetes pods with specialized hardware like GPUs, inspired by talks at KubeCon NA 2024

CuTeDSL on Hopper - Pipelining (veitner​.bearblog​.dev). Exploring Hopper's kernel methods, TMA atoms, pipeline techniques, and memory management for efficient GMEM to SMEM transfers in CuTeDSL

🔧 Data Engineering & Systems

CS Researchers Win Top Honors at SIGMOD 2025 (cs​.columbia​.edu). Columbia University’s Data Management Group showcased its research at SIGMOD 2025, winning awards and presenting on data lineage, visualization, and machine learning techniques

Unlocking Unstructured Data with LLMs (thedataexchange​.media). Shreya Shankar discusses LLMs and DocETL for processing unstructured data, semantic extraction, thematic analysis, and enterprise applications in the Data Exchange podcast

Robust Time Series Monitoring: Anomaly Detection Using Matrix Profile and Prophet (blog​.sentry​.io). Matrix Profile and Meta's Prophet enhance time series anomaly detection by addressing noisy metrics and evolving patterns for robust production system monitoring

Xiao Hu and colleagues receive 2025 SIGMOD Research Highlight Award (uwaterloo​.ca). Xiao Hu and colleagues awarded 2025 SIGMOD Research Highlight for 'Reservoir Sampling over Joins', presenting a new efficient sampling algorithm for database queries

Data Engineering Best Practices: Building Scalable and Reliable Data Pipelines (ataiva​.com). Explore data engineering best practices for scalable, reliable data pipelines, covering architecture, ETL/ELT processes, data quality, governance, and tools like Apache Airflow and dbt

⚙️ Implementation & Development

Adding client-side semantic search to jonmagic.com (jonmagic​.com). Implementing client-side semantic search on a static site using MiniLM-L6-v2 embeddings, cosine similarity, and JavaScript without a backend

Debugging onnxruntime-web models (a2p​.it). Debugging onnxruntime-web models using techniques like inspecting intermediate tensor shapes, custom builds, and intercepting console logs

Experimentation over capabilities (efekarakus​.com). Organizations adopt 'capability' in generative AI, emphasizing experimentation over rigid feature delivery to enhance evaluation and performance of LLM systems

Calling =TECHTO_MLREGRESSION for Machine Learning supervised regression in Excel is just a matter of copying and pasting (thierrymoudiki​.github​.io). Use =TECHTO_MLREGRESSION in Excel for Machine Learning regression with xlwings Lite and Python integration. Get started with a simple installation and token setup

🎯 Model Training & Optimization

How to Monitor, Diagnose, and Solve Gradient Issues in Foundation Models (neptune​.ai). Explore gradient issues during foundation model training, utilizing tools like Neptune.ai for real-time monitoring and applying solutions like gradient clipping

Can tinygrad win? (geohot​.github​.io). tinygrad seeks to redefine software development for machine learning, aiming for speed and efficiency with a minimal number of lines of code

Microsoft’s Reinforcement Pre-Training (RPT) – A New Direction in LLM Training? (aipapersacademy​.com). Microsoft's Reinforcement Pre-Training (RPT) enhances large language models using next-token reasoning and Group Relative Policy Optimization to scale reinforcement learning

Per-Tensor and Per-Block Scaling Strategies for Effective FP8 Training (developer​.nvidia​.com). Explores FP8 scaling strategies including per-tensor, delayed, current, and per-block scaling for effective numerical stability in low-precision training, utilizing NVIDIA's tools

📐 Mathematics & Theoretical Foundations

Beginning at the Beginning (eigenvector​.com). Foundation in linear algebra essential for data science; Eigenvector offers courses linking theory to practice, emphasizing multivariate methods and modeling concepts

What is a Good Quantum Encoding? Part 1 (math3ma​.com). Exploration of quantum machine learning, encoding classical data onto quantum computers, category theory, and challenges in quantum feature mapping by researchers from MIT and Deloitte

PCA analysis of Futures returns for fun and profit, part #1 (qoppac​.blogspot​.com). Explore PCA analysis of futures returns, uncover risk factors, trading strategies, and the relationship between risk and return in systematic trading

Copilot in Excel: How to do principal component analysis with Python (stringfestanalytics​.com). Learn to perform Principal Component Analysis (PCA) using Python in Excel, simplifying complex datasets into meaningful components with Copilot support

A non-anthropomorphized view of LLMs (addxorrol​.blogspot​.com). Exploration of LLMs as mathematical constructs, contesting anthropomorphism and discussing AI alignment, safety, and the nature of human consciousness

Measuring Tree Similarity (gojiberries​.io). Explore tree similarity measurement techniques, focusing on interpretive stability, partition stability, and predictive stability using tools like cophenetic correlation and variable importance

Atoms of shape, or On the origin of doodling (logarithmichistory​.wordpress​.com). Explores geometric operations that generate shapes, linking language and thought in humans through primitive geometry and cultural history

The Broken Mirror: What Generative Models Still Don’t Understand About Symmetry (riccardo-disipio​.medium​.com). Exploring generative models and their struggles with mirror symmetry, highlighting insights from physics and the role of Emmy Noether's theorem in understanding structure

👋 Before you go

I've got a big favor to ask - keeping Blaze running isn't expensive, but it does all add up, so I'm asking readers like you to help, if you can.
That's why I'm launching a Patreon page!. Nothing flashy, just a way for folks who find value in these newsletters to chip in a little each month. In return, you'll get:

  • Real say in how Blaze evolves — vote on new topics, features, topic curation ideas
  • First dibs on merch (details still cooking)
  • That warm fuzzy feeling knowing you're supporting something that saves you time and keeps you plugged into great tech writing

If you are getting value from blaze, checking this out would mean the world. And if you can't contribute, no worries—the newsletters keep coming either way, and you can follow along on patreon for free.
Thanks for reading and being part of this nerdy corner of the internet. All the best - Alastair.

You may also like

About Machine Learning Engineer

Our Machine Learning Engineer newsletter covers the latest developments, research papers, tools, and techniques in ML engineering and deployment. Each week, we curate the most important content so you don't have to spend hours searching.

Whether you're a beginner or expert in machine learning engineering, our newsletter provides valuable information to keep you informed and ahead of the curve in this technically challenging field.

Subscribe now to join thousands of professionals who receive our weekly updates!