字节跳动
校招新加坡 · 校招 · 2027 届 · 博士
长期有效
团队介绍:字节跳动数据中心与供应链部门,负责字节跳动从芯片到服务器全栈算力基础设施的研发、调度与供给、管理与应用(如数据中心建设、内核操作系统、网络CDN、高性能服务器、编译工具链等),支撑字节跳动全系业务运行的核心力量,为抖音、今日头条、飞书、火山引擎等各类产品提供全面、领先、稳定的百万量级大规模算力基础设施服务。
课题介绍:随着智能计算与AIDC产业快速发展,数据中心单机柜功率密度持续攀升,传统基础设施在换热效率、节水效果、能源利用方面存在短板,同时海量运维数据未充分利用,运维效率依赖人工经验,难以适配产业高质量发展与政策要求。
本课题从技术创新与智能化赋能两方面开展研究:一方面围绕液冷技术、节水无水冷源、供配电储能方向协同创新,突破现有技术瓶颈,适配高密度功率需求与双碳、节水政策要求;另一方面构建数据中心运维AI Agent体系,利用大模型技术从海量多模态异构运维数据中学习,实现多Agent协作完成从监控、诊断到修复的全流程自动化,打造PUE智能优化闭环,支撑从"被动响应"到"主动预测与自愈"的范式转变。通过硬件技术创新与AI智能化运维深度结合,全面提升数据中心能效、可靠性与运维效率。
Topic Content
As intelligent computing and the AIDC industry develop rapidly, data center rack power density continues to rise. Traditional infrastructure struggles with heat exchange efficiency, water-saving, and energy use. At the same time, large amounts of maintenance data are underused, and maintenance relies heavily on human experience, making it hard to meet the industry's quality and policy demands. This topic focuses on two main areas: technology innovation and smart operation. First, it aims to innovate in liquid cooling, water-saving or water-free cooling, and power supply and energy storage to overcome current limits, meet high-density power needs, and comply with carbon reduction and water-saving policies. Second, it builds an AI agent system for data center maintenance that uses large AI models to learn from vast, varied multimodal maintenance data. This system enables multiple AI agents to work together to automate everything from monitoring and diagnosis to repair, creating an intelligent loop to optimize the power usage effectiveness (PUE) and shifting maintenance from "reacting after problems" to "predicting and fixing them automatically". By combining new hardware technology with AI-driven maintenance, the project seeks to improve data center energy efficiency, reliability, and operation efficiency.
Topic Challenges
Topic Value
上传简历,AI 会对照这份 JD 指出你的差距和需要补强的地方
基于 算法 岗位的常见面试问题整理
请解释 Transformer 模型的自注意力机制(Self-Attention)
1. Self-Attention = 序列中每个 token 与所有 token 计算相关性权重 2. Q=Wq·x, K=Wk·x, V=Wv·x; Attention(Q,K,V)=softmax(QK^T/√d_k)V 3. 除以 √d_k 防止点积过大导致 softmax 梯度消失 4. 多头注意力:多组 QKV 并行,拼接后线性变换,捕获不同子空间特征 5. 对比 RNN:可并行计算,捕获长距离依赖
你做过的最有深度的算法/模型项目是什么
1. 一句话说清任务:分类/生成/检索/推荐 2. 模型选型:为什么选这个而非那个 3. 数据:数据量、来源、清洗策略 4. 创新点:你做的改进是什么(不是调参) 5. 结果:核心指标提升(准确率/召回/延迟)
过拟合是什么?怎么判断和解决
1. 过拟合=训练集表现好、验证集表现差,模型记住了噪声 2. 判断:train loss 持续下降而 val loss 上升 3. 解决:更多数据 / 正则化(L1/L2) / Dropout / 早停 / 数据增强 / 降低模型复杂度
解释梯度下降的原理,SGD 和 Adam 的区别
1. 梯度下降:沿负梯度方向更新参数,使损失降低 2. SGD:每次用随机小批量估算梯度,引入随机性帮助跳出局部最优 3. Adam = Momentum + RMSProp:自适应学习率 + 动量加速 4. Adam 收敛快但可能泛化不如 SGD;CV 常用 SGD,NLP 多用 Adam
给你 10 万条未标注的文本数据,如何找出其中 10 个主题
1. 预处理:分词、去停用词 2. 向量化:TF-IDF 或 Sentence Embedding 3. 聚类:K-Means(需指定 K=10)或 HDBSCAN(自动确定类数) 4. 或 LDA 主题模型直接输出主题-词分布 5. 评估:看每个簇的高频词判断主题合理性,迭代调整
更多面试题库功能即将上线
信息来源:企业官方招聘页。投递前请以官网信息为准。