字节跳动
校招大模型时代下的搜广推系统创新研究-Data-抖音/直播/电商/剪映(新加坡)Innovative Research on Search, Advertising, and Recommendation Systems in the Era of LLMs-Data-TikTok/Live/E-commerce/CapCut (Singapore)
新加坡 · 校招 · 2027 届 · 博士
长期有效
任职要求
- 2027届毕业,获得博士学位,人工智能、计算机科学、计算机工程及相关技术专业
- 具备优秀的编程能力,熟练掌握数据结构与基础算法;至少掌握C/C++/Python中一种语言,同时具备扎实的工程思维,能够兼顾性能与成本的平衡。
- 加分项:
- 在VLDB、SIGIR、OSDI等CCF A类期刊或会议有论文发表者优先
- 有韧性,具备良好的沟通能力与团队协作能力;热爱技术,愿意与团队共同迎接挑战,富有创新精神。
- Individuals who are completing or recently completed a PhD in Artificial Intelligence, Computer Science, Computer Engineering, or a related technical discipline
- Excellent programming abilities with a strong command of data structures and fundamental algorithms; For traditional coding roles, proficiency in C/C++ is required; for intelligent coding roles, proficiency in Python is required; Candidates are required to use these languages to implement complex algorithms and build iterative models; Candidates should also have a strong engineering mindset with the ability to balance performance and cost.
- Preferred Qualifications
- Candidates with publications in CCF-A category journals or conferences such as VLDB, SIGIR, OSDI, etc., are preferred
- Strong resilience, excellent communication and teamwork skills; passionate about technology, willing to embrace challenges with the team, and a drive for innovation.
岗位描述
团队介绍:推荐架构团队支撑字节跳动旗下多款APP产品,如抖音、今日头条、番茄小说、西瓜视频、剪映等推荐系统架构的设计和开发,保障系统的稳定和高可用,致力于抽象系统通用组件和服务,建设推荐中台、数据中台;关于在线服务,在这里你有机会参与大规模机器学习在线预估框架的研发与优化,也有机会参与模型训练与调度等相关问题的研究与突破,解决系统瓶颈,降低成本开销;如你对大数据感兴趣,在这里也有机会参与通用实时计算系统的开发、构建统一的推荐特征中台,为推荐业务实现先进的消重、计数、特征服务等;我们期待热爱技术的你加入,一起创造更多可能。
课题介绍:随着业务场景日趋复杂,推荐/广告/搜索面临严峻挑战。大模型虽能精准捕捉用户偏好、提升个性化效果及内容质量,但实时性、稳定性和可扩展性要求极高,在分布式训练、推理加速、异构硬件及多模态数据处理等方面存在诸多难点。同时,模型规模扩大与多模态数据涌现,使现有基础设施在数据处理效率与资源利用上难以满足需求。
本课题聚焦系统与工程创新,突破关键技术瓶颈,构建高效、稳定、可扩展的大模型解决方案,为搜广推场景提供坚实技术支撑。
课题挑战
- 1、大模型原生训推架构重构;
- 2、极致性能优化与AI Infra创新;
- 3、端到端生成式范式创新;
- 4、多模态AU数据基础设施与质量优化;
- 5、多模态数据表征和 RAG 应用系统。
课题价值
为搜广推业务构建下一代生成式AI基础设施,通过大模型、多模态技术与系统协同创新,突破性能瓶颈,实现超长上下文、毫秒级响应和极致信息精度,驱动业务智能升级。
Topic Content
As business scenarios become increasingly complex, search, advertising, and recommendation are facing significant challenges. While large models can accurately capture user preferences and enhance personalization as well as content quality, they also impose stringent requirements on real-time performance, stability, and scalability. This introduces substantial technical challenges in areas such as distributed training, inference acceleration, heterogeneous hardware utilization, and multimodal data processing.
At the same time, the rapid growth in model scale and the proliferation of multimodal data have made it difficult for existing infrastructure to meet the demands of data processing efficiency and resource utilization. This topic focuses on system and engineering innovations to overcome key technical bottlenecks and build efficient, stable, and scalable large-model solutions, providing a robust technical foundation for search, advertising, and recommendation scenarios.
Topic Challenges
- 1.Native training and inference architecture redesign for LLMs;
- 2.Extreme performance optimization and AI infrastructure innovation;
- 3.End-to-End generative paradigm innovation;
- 4.Multimodal AU data infrastructure and quality optimization;
- 5.Multimodal data representation and RAG-based application system.
Topic Value
Building next-generation generative AI infrastructure for search, advertising, and recommendation businesses. Through the co-design of large models, multimodal technologies, and system-level innovations, we aim to overcome performance bottlenecks and enable ultra-long context handling, millisecond-level response latency, and high-precision information understanding, thereby driving intelligent upgrades across the business.
你的简历匹配这个岗位吗?
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
该岗位可能会问的问题
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)
查看答题思路
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
查看答题思路
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
查看答题思路
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
查看答题思路
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
查看答题思路
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。