0 views collected around this technical thread.
This article explores essential sampling strategies for big‑data environments—including simple random, reservoir, stratified, oversampling, undersampling, and weighted sampling—detailing their principles, algorithmic steps, advantages, drawbacks, and suitable application scenarios to help analysts choose the right method.
The article examines the performance impact of excessive logging in large‑scale Java systems and proposes request‑level sampling with cross‑thread identifier propagation, offering practical component‑based solutions, implementation considerations, and a concrete code example for backend developers.
The article reviews the development of online A/B testing, covering sampling and traffic‑splitting techniques, metric computation improvements, statistical inference advances, and current challenges such as interference, real‑time inference, and large‑scale metric computation, while referencing recent research papers.
This guide introduces the fundamentals of AI image generation with Stable Diffusion, covering three main usage methods, the Draw Things desktop app, model types, samplers, prompts, and post‑processing techniques to create high‑quality promotional graphics for events like the 618 sale.
The article explains how greedy search, beam search, Top‑K, Top‑p (nucleus) sampling, and temperature each shape large language model generation, comparing their effects on repetition, diversity, and creativity, and provides concise TensorFlow‑based code examples illustrating these inference‑time strategies.
The article explains how cloud‑native observability systems gather massive telemetry from infrastructure, containers, middleware and services, compares direct push and file‑based collection approaches, and details head, tail and local sampling methods to optimize data completeness and performance.
The article explains that while observability is not new, cloud‑native systems have driven rapid development of observable platforms, detailing data collection architectures, direct push versus file‑based approaches, and various sampling techniques (head, tail, and local sampling) to balance completeness, real‑time reporting, and performance impact.
The article explains how generative models use deterministic methods like greedy and beam search and stochastic techniques such as top‑k, top‑p, contrastive search and sampling, describing their mechanisms, temperature control, repetition penalties, and practical trade‑offs for balancing fluency, diversity and coherence.
The paper introduces a pre‑user stratification model that uses covariate‑balancing algorithms to create separate strata for distribution and revenue metrics, ensuring equal user allocation in Vivo game‑center AB tests, which reduces metric variance, improves gray‑release effectiveness, and saves significant investigation effort.
This article explains how Ant Group tackles the performance challenges of large‑scale OLAP tables by using ADM to reduce data volume, employing Cube pre‑aggregation for reporting, and applying statistical sampling for exploratory analysis, detailing the processes, metrics, and architectural designs involved.
This tutorial explores various text generation decoding methods—including greedy search, beam search, top‑k/top‑p sampling, sample‑and‑rank, and group beam search—explaining their principles, providing detailed Python code examples, and comparing their use in modern large language models.
This guide thoroughly explains the major decoding strategies for neural text generation in HuggingFace Transformers—including greedy, beam, diverse beam, sampling, top‑k, top‑p, sample‑and‑rank, beam sampling, and group beam search—detailing their principles, Python implementations with LogitsProcessor components, workflow diagrams, comparative analysis, and references to original research.
This article describes ByteDance's comprehensive approach to managing the massive volume of event‑tracking (埋点) data, detailing the background, cost‑reduction strategies, experience review, future plans, and a Q&A session that together illustrate how systematic data governance can dramatically cut storage and processing expenses.
This article explains ByteDance's large‑scale adoption of profile‑guided optimization (PGO), covering its principles, instrumentation and sampling methods, the automated platform built for data collection and compilation, and the resulting performance gains across dozens of critical services.
This article explains the fundamentals of Markov chains, illustrates their transition matrix with a market example, demonstrates convergence through Python code, and outlines how to use the stationary distribution for sampling in Monte Carlo simulations.
This article explains the detailed‑balance condition for Markov chains, shows why finding a transition matrix for a given stationary distribution is hard, and demonstrates how Metropolis‑Hastings modifies MCMC to achieve higher acceptance rates with a concrete Python example.
This article explains the basic definition of Markov chains, illustrates a stock‑market example with transition matrices, demonstrates convergence through Python simulations, and shows how the steady‑state distribution enables sampling for Monte Carlo methods.
This article introduces Monte Carlo methods, explains how random sampling approximates integrals, discusses uniform and non‑uniform probability distributions, and details acceptance‑rejection sampling as a technique for generating samples from complex distributions, laying the groundwork for understanding Markov Chain Monte Carlo in AI.
This article explains the motivations behind Monte Carlo methods, introduces acceptance-rejection sampling, details Markov Chain Monte Carlo concepts, and walks through Metropolis-Hastings and Gibbs sampling algorithms with Python implementations, highlighting their use in high‑dimensional probability distribution sampling.
Monte Carlo methods, originally a gambling-inspired random simulation technique, provide a versatile way to approximate integrals and sums, and by using acceptance‑rejection sampling they enable drawing samples from complex probability distributions, a key step toward effective Markov Chain Monte Carlo algorithms in machine learning and AI.