Peter Song

Peter Song is a machine learning, data engineer, and AI writer with a focus on practical ML workflows. He shares insights on machine learning, data tools, and MLOps.

How to Reduce Hallucination in LLM Applications

by

Hallucination—when large language models confidently generate plausible-sounding but factually incorrect information—represents one of the most critical challenges preventing widespread adoption of LLM applications in high-stakes domains. A customer support chatbot inventing product features, a medical assistant citing nonexistent research studies, or a legal research tool fabricating case precedents can cause serious harm to users and … Read more

Convolutional Neural Network Architectures for Small Datasets

by

Deep learning’s most celebrated successes—ImageNet classification, object detection, semantic segmentation—share a common ingredient: massive datasets with millions of labeled examples. ResNet trained on 1.2 million images. BERT consumed billions of words. Yet most real-world computer vision problems don’t come with millions of labeled images. Medical imaging datasets might have hundreds of scans. Manufacturing defect detection … Read more

Memory-Efficient Attention Algorithms: Flash Attention, xFormers, and Beyond

by

The attention mechanism sits at the heart of modern transformers, enabling models to weigh the importance of different input elements when processing sequences. Yet this powerful mechanism comes with a significant cost: memory consumption that scales quadratically with sequence length. For a sequence of 8,192 tokens, standard attention requires storing an 8,192 × 8,192 attention … Read more

Graph-Based ML Algorithms vs Graph Neural Networks

by

Graphs are everywhere in our data-driven world. Social networks connect billions of users, molecules are represented as atoms connected by bonds, transportation systems link cities through roads and railways, and knowledge graphs organize information through relationships. When it comes to extracting insights from these graph-structured datasets, practitioners have two fundamentally different approaches at their disposal: … Read more

Online Learning Algorithms for Streaming Data: Adapting in Real-Time

by

In an era where data flows continuously from countless sources—social media feeds, financial markets, IoT sensors, user interactions, and network traffic—the traditional batch learning paradigm struggles to keep pace. Batch learning assumes you can collect all your data, train a model once (or periodically retrain), and deploy it until the next training cycle. But what … Read more

Large Margin Classifiers Beyond SVMs

by

Support Vector Machines (SVMs) have long been synonymous with large margin classification. Their elegant mathematical foundation and proven effectiveness made them the go-to choice for practitioners seeking classifiers that maximize the separation between classes. Yet the concept of large margin learning extends far beyond SVMs, encompassing a rich family of algorithms that apply margin-based principles … Read more

Facebook Prophet vs Classical ARIMA vs LSTM

by

Time series forecasting remains one of the most practical and widely deployed machine learning applications. From predicting stock prices and sales volumes to forecasting energy consumption and website traffic, the ability to anticipate future values based on historical patterns drives critical business decisions. Yet choosing the right forecasting method can feel overwhelming—should you use the … Read more

How Decision Trees Choose Split Points Using Gini Impurity vs Entropy

by

Decision trees stand as one of the most intuitive and widely-used machine learning algorithms, making complex decisions through a series of simple yes-or-no questions. At the heart of every decision tree lies a critical challenge: how to determine the best way to split data at each node. This seemingly simple question has profound implications for … Read more

Custom Model Deployment with SageMaker Endpoints

by

Deploying machine learning models to production is one of the most critical yet challenging phases of any ML project. While training a model that achieves excellent accuracy on test data is an accomplishment, the real value emerges only when that model serves predictions reliably at scale. Amazon SageMaker Endpoints provide a powerful managed infrastructure for … Read more

Building Low-Latency Inference APIs Using FastAPI and ONNX

by

Latency kills user experience and revenue. In production ML systems, every millisecond of inference delay compounds across millions of requests—a model taking 200ms instead of 50ms doesn’t just slow down four requests, it reduces your system’s throughput capacity by 75% and degrades user experience enough to measurably impact conversion rates. Whether you’re serving recommendations that … Read more