Internship Reflections 实习心得体会
Published:
Half a year since my internship ended, I’ve gained new perspectives on the challenges and thoughts I faced when first entering the workplace. Here’s a summary looking back.
The Team
I chose the team of the interviewer who actively recruited me — the federated model algorithm team at JD Technology’s Risk Management Department. The team was small, but the leader Hu Wang had clear vision, leveraging years of industry experience to establish the federated learning direction and lead the team to breakthrough results while advancing engineering through collaborative projects.
How to Quickly Adapt to a New Research Direction
Federated learning was new territory for me, and the short internship timeline made rapid learning essential.
My approach:
- Leverage existing expertise — learned from a senior colleague’s survey papers, latest breakthroughs, and proposed research directions
- Iterate based on new requirements — experimented with different model frameworks, constantly communicating and adjusting
- Stay in close dialogue — frequent exchanges to prevent going too far off track
How Research Teams Survive in Industry
This team was a frontier research group within a business department — its positioning differed significantly from the core business. I later learned that after the team leader left, the direction shifted toward supporting the main business.
The pressure is universal: wanting both business impact and cutting-edge research, but struggling to quickly commercialize frontier work. I’ve heard comments like “By the time our research team ships something usable, we could’ve bought Company X’s product.”
A rare success story: Tencent’s vision lab leveraged gaming scenarios to find incredible applications for their research — though the first 3-5 years of research had no consideration for deployment at all.
Managing Work Pressure
During the process of proposing new model frameworks, since predecessors already had a rough direction, I was essentially guided toward fixed approaches. This was painful — constant rejection of my own thinking, followed by accepting others’ ideas. The lack of creative freedom and validation was unprecedented pressure.
My coping strategy: leaving work earlier for self-recovery, and extensive communication to directly understand predecessors’ thinking rather than over-thinking on my own.
Different personalities suit different leadership styles. The simpler approach is, of course, choosing a leader that fits you.
实习结束近半年，对于刚入职场时面对的问题、想法有了很多变化，从未来看过去，总结一下。
实习团队
选择坚持找我面试的面试官团队——京东科技风险管理部门做联邦模型算法的团队。团队不大，leader Hu Wang 很有规划，带领团队在联邦学习算法上重点突破，同时借助合作项目带动工程进展。
如何快速融入新方向
联邦学习是此前没深入学习过的内容，加上实习时间短，挑战很大。
方法：
- 借着师兄的肩膀，学习综述文章和最新突破
- 根据新场景要求尝试调整不同模型框架
- 不断交流调整思路，防止走偏
企业研究团队如何发展
这个团队属于业务部门中的前沿算法研究团队，定位与主营业务差别较大。后来团队Leader离开后，方向转向支撑主营业务。
压力相同：既想取得业务收益，又无法快速将前沿技术落地。”等我们研究团队落地实用化，还不如花钱买XXX公司的产品”。
成功案例：腾讯视觉实验室借助游戏场景取得了极好的落地，但此前3-5年研究时是没有考虑落地的。
控制工作压力
提出新模型框架的过程中，因前人已有大致方向，基本是拧着我走。大量思考被否定，然后接受别人的思路，没有自由度和认可带来的压力前所未有。
应对：通过早下班自我排解+大量沟通直接理解前人想法，减少自己过度的想法。
不同性格适合不同领导风格，更简单的方式是选择适合自己的领导。