Major Internet companies are facing a 'wave of layoffs'. How can algorithm engineers survive the 'cold winter' of their careers?

1. Three important stages in the workplace

About career path Thinking involves a philosophical question: Where do I come from and where do I want to go? As an algorithm engineer, you will generally face the following three stages of challenges in your career.

1. Challenge 1: Transition from student to workplace

This is the only way for all working people . In school, the relationship between tutors and students is relatively simple and simple; in the workplace, you will encounter many complex things that you have not been exposed to before. In this regard, there are several suggestions as follows:

① Participate in at least one internship before officially joining the job to adapt to workplace life in advance.

② Stay curious in the workplace, consult and communicate with senior people more, talk less, watch more and do more.

③ Maintain learning ability in the workplace, focus on the accumulation of basic knowledge and abilities, and combine them with practical problems in the workplace.

④ Correct your mentality and don’t dislike dirty work. You can also accumulate experience and credit in small things.

2. Challenge 2: Transition from Junior to Senior

This is also the growth experienced by most senior professionals path. Taking major domestic manufacturers as an example, Tencent’s Junior ranks are approximately 6-9, which is roughly equivalent to Alibaba’s P5-P7, corresponding to the growth process from Junior Engineer to Senior Engineer. This stage is an important stage for accumulating business skills and technical depth. The length of time each person goes through in this stage varies, depending on career opportunities, personal status, and business growth space. In this regard, there are several suggestions:

(1) Deeply understand the nature of the business​

Compared with wild ideas and research, the "implementation" of ideas is more important. The recent cutting-edge development directions of the Internet industry, including the emergence of large models such as Chat-GPT and AIGC, are the product of technological innovation that closely follows user needs and corresponding products. The large language model behind the recently popular Chat-GPT actually appeared very early. However, due to the lack of application "tipping point" in the early stage, it was not able to be "implemented" through the product, so it has not been widely used. As an algorithm engineer, you must start from the nature of the product and business to understand and tap the value brought by the algorithm.

(2) Deeply dig into technology​

Continuously expand and improve your own technology ability. Taking the recommendation system as an example, it is necessary to have a full-stack understanding of the modeling capabilities of models such as recall and sorting (including mixed ranking and fine ranking), as well as the advantages and disadvantages of SOTA models, and the progress of cutting-edge models.

(3) Development of T-shaped talents

T-shaped talents are often The concept mentioned refers to improving one's technical scalability while ensuring technical depth in one's own technical track. Scalability includes two aspects, one is the business level and the other is the technical level. Still taking the recommendation system as an example, algorithm engineers in the recommendation system direction recommend learning more about the basic principles of CV and NLP to better empower related businesses.

In general, the work of Junior engineers is more focused on task execution, while the work of Senior engineers is more focused on thinking and solving specific problems. For example, if the DAU of a certain APP declines, you need to find out the reasons for the decline and propose solutions. This is a big topic. It may be caused by a relatively low conversion rate of a certain page or a certain scene, or it may be caused by various other complex reasons. This is an issue that a Senior engineer needs to consider systematically. First, identify key issues through data analysis, then set goals, build models, construct evaluation indicators, then propose solutions, and finally solve the problems. In such a process, it is often difficult to achieve the goal by individuals, and it is necessary to coordinate various external resources to complete it.

3. Challenge 3: Transition from employee to Leader

The third challenge, some engineers may have I have experienced it, but most engineers may not have experienced it yet, and that is the transition from employee to leader. This is a big leap, because the leader needs to lead the team to achieve goals, so the responsibility is even greater. On the other hand, the management model of Internet companies is gradually becoming flatter, so there are not many positions left for leaders. If you are lucky enough to become a leader, you must first believe in yourself and boldly lead the team to achieve its goals. When you first become a leader, you often face a difficult problem, which is the balance between business and technology. This involves the art of management. The essence of management is actually inseparable from three major elements: responsibility, power, and money. "Responsibility" represents the division and definition of responsibilities; "right" represents the arrangement of personnel's work; "money" represents the incentive mechanism for employees.

The transition from employee to leader is a very challenging job. With the growth of experience and the improvement of communication skills, these abilities will improve. formed imperceptibly. In addition, as a leader, you need to constantly expand outward and mobilize the resources of other teams to the greatest extent, instead of just asking or even squeezing from within.

The above are the three stages of challenges that algorithm engineers often face. Each stage has its own thorny problems, but there are also corresponding solutions. . As an algorithm engineer, you must have enough patience at every stage and be calm and polish yourself.

2. How do algorithm engineers plan?

There are 3 key points in career planning for algorithm engineers ：

① Vision: You must see the situation clearly before making plans, otherwise incorrect plans may be made.

② Self Evaluation: Before planning, you must conduct a comprehensive self-evaluation of yourself, "know yourself and the enemy, and you will never be in danger", so as to choose the direction that suits you for planning;

③ Action: Take action! No matter how perfect the planning is, it is not as practical as taking action.

1. Vision

To control the situation, you need to do the following:

① First of all, we must ensure that we can see clearly and understand the current situation clearly.

② Secondly, we must ensure that we see the whole picture and examine the prospects of the current industry from a more comprehensive perspective.

③ Finally, make sure to see far. The "far" mentioned here not only refers to the scope, but also the time span; only the farther time can be seen Only through cycles can we plan for the future more clearly.

(1) National Games Level​

The above picture is taken from the book "Principles" by Ray Dalio, a famous American venture capital expert. The author in the book established a model to measure the rise and fall of empires. The curve in the above picture reflects The rise and fall index of empires over time. The blue curve in the picture represents the national destiny of the United States, which is highly consistent with the actual situation: in the 1950s, the national destiny of the United States reached a peak and achieved many technological breakthroughs; in contrast, China ( The red curve in the figure), when liberation had just been achieved, the country was impoverished and at a low point; the timeline was further advanced to 1500 (around the Ming Dynasty), when China was in a leading position in the world; in modern times, China China's empire index has always been at a low level; by 1950, China began to develop rapidly; until now, China's empire index has gradually approached that of the United States, while the United States is declining.

# Of course, different experts will build different models and have different understandings based on their own understanding. In the same way, when carrying out industry vision, you must also combine your own cognition with the understanding of experts to conduct a comprehensive and comprehensive analysis.

(2) Mobile Internet industry level​

Specific to the Internet level, especially It is at the mobile Internet industry level. You can refer to the figure below. The chart below shows the monthly active user base from QuestMobile.

#As can be seen from the picture, in the past In the past three years, the entire Internet has not experienced major growth. The annual net DAU growth is only 20 million (a certain product has more than 100 million DAU). Therefore, this level of growth is difficult to support the growth of an APP. From this point of view, the scale of Internet users has long been stable. Therefore, the earlier method of using demographic dividends and gaining profits through crowd tactics is no longer feasible, and the entire Internet industry has become a stock market. This is something that needs to be recognized now. Reality.

The Internet industry is further subdivided into the following tracks:

(1) E-commerce

① In the early years, e-commerce companies such as Pinduoduo, Alibaba, and JD.com used the crowd-sea tactic and used subsidies to achieve large-scale development.

② In recent years, the space for user growth has become smaller and smaller, and the demographic dividend has gradually disappeared, so similar subsidies will become less and less; with Double Eleven For example, in recent years, e-commerce shopping companies no longer pursue the GMV transaction volume of the day, but pursue maximization of profits more rationally.

③ The only demographic dividend in the e-commerce industry in the near future may come from the sinking market, but the growth space is still limited.

④ Therefore, , the future development direction of e-commerce will be towards quality e-commerce and vertical e-commerce.

(2) Community

① The community has developed rapidly in recent years; taking Xiaohongshu as an example, the community atmosphere of Xiaohongshu is very good: Users are constantly "planted" by the content of Xiaohongshu and formed their minds; then users will actively participate in relevant topic discussions and resonate with each other, thus forming revenue conversion.

② The community is a developing track, and we are optimistic about its future development; for some niche vertical communities, although they are not large in scale, their quality is very high. high.

③ Community development does not rely on demographic dividend, but more on penetration rather than crazy growth, so it is a promising development direction.

(3) Game

##① The prospect of the game track is relatively promising. In addition to the continued growth of the domestic game business, there is also a big gap in the game track. The blue ocean lies in overseas business. Many domestic game operation experiences can be transplanted overseas. This is also a recent business direction of Tencent.

② Generally speaking, the game track is not affected by the demographic dividend of the Internet, and at the same time it has broad overseas development space. It is a A relatively promising track.

(4) Social Network

① The number of WeChat users is close to the number of domestic Internet users and population, which is a very large scale.

② The moat of social networks is very deep. It is difficult for users to easily migrate from the original social platform to another social platform because the cost of involvement is high. This is also One of the reasons why ByteDance has tried many times in the social networking business but has never been successful.

③ Although social networks will be affected by the demographic dividend to a certain extent, due to their high barriers, there is a high probability that they will maintain a stable status quo in the future, that is, With WeChat as the core, various extensions based on the WeChat ecosystem are gradually formed.

④ The algorithms involved in social networks are mainly based on graph and community propagation methods. Such methods bring very limited value to small-scale social networks. Only for social networks of the size of WeChat can algorithms such as graph models, social network communication chains, and community discovery be of corresponding value.

⑤ In summary, the development of social networks will be relatively stable, and it is difficult to have big opportunities in the short term.

(5) Information Platform

① Information platforms have also been declining recently, and users do rely on information platforms to a certain extent. , but the degree of dependence is not strong.

② In the past two years, information platforms have been greatly impacted by short videos, which have captured a lot of the user market.

③ In this context, information platforms will return to their essence, which is to distribute information and meet users’ information needs in specific fields.

④ For recreation, entertainment, kill-time and other long-tail information, users generally obtain it through short video platforms, which brings new challenges to information platforms. A bigger challenge.

⑤ Due to the complex medium of the information platform, numerous rules, and strict supervision, it is difficult for users to adjust their opinions. In addition, coupled with the impact of the external short video field, the information platform’s The difficulty is "next level".

⑥ In addition, information platforms also rely more on demographic dividends. Nowadays, demographic dividends tend to be saturated, further restricting the development of information platforms.

⑦ In summary, the choice of information platform track should be cautious.

(6) Short video

① The recent development of short video is booming, with the longest user time, the richest user behavior, and the most positive users. Product forms with the most intensive negative feedback are the most frequent.

② The short video track has rich data types and a huge number of users, so the potential value of the data is high and there is huge room for future development.

③ Recently, short video advertisements and even live broadcasts have become increasingly common. These signs indicate that short videos are gradually being integrated with e-commerce models and have huge potential for monetization.

④ In summary, short video is a track with many opportunities and great potential.

(3) AI industry level​

##The above picture shows the changing trend of the life cycle form of AI technology over time:

① It is in the climbing stage of the left curve It is an emerging AI technology, and the technological prospects remain to be seen.

② At the trough of the middle curve are AI technologies with uncertainty. These technologies still need time and market testing. A considerable number of AI technologies will Facing the "bubble burst".

③ The curve on the right shows that AI technology has broken through the "bubble burst" and demonstrated and precipitated the value of AI technology. If combined with Better user needs and product applications will lead to a "comeback" for this technology.

④ The right end of the curve is the most ideal stage of AI technology development, and the productization of technology continues to bring objective growth and income.

Take several recent popular AI directions as examples to introduce in detail:

(1 )AIGC

#AIGC is very popular recently, such as Stable Diffusion, Midjourney and other AI painting tools. The emergence of Chat-GPT has also subverted many models in the CV and NLP fields, reflecting the strong product power of large models. Many engineers worry that the emergence of such large models will pose a threat to algorithm engineers, and even to all mankind. In fact, it is still far from a "threat", and the realization of general artificial intelligence (AGI) still has a long way to go. , whether it is AI painting or ChatGPT, the algorithm does not yet have logic and consciousness capabilities. When a technology develops to a certain stage and encounters a good product idea, it will definitely explode. Therefore, whether it is AI painting or ChatGPT, there are user needs and product ideas as support behind it. The enlightenment brought by this is: technology and business will never be separated. Only by fully understanding the business can the value brought by algorithm technology be maximized.

(2) Autonomous driving

Autonomous driving is a promising direction, aiming to solve an ultimate problem: how to optimize road traffic flow. Autonomous driving is divided into many levels of goals. The first thing to be achieved is "bicycle intelligence". At present, many companies have achieved satisfactory results, such as Tesla Autopilot. Abroad, FSD autonomous driving has reached a relatively mature stage; in China, the autonomous driving technology of first-tier autonomous driving companies such as Xpeng, an up-and-coming car company, and Baidu Apollo, an independent third-party platform, is also becoming increasingly mature. Autonomous driving itself is a product that assists or even replaces humans, liberating human labor to a certain extent and making people feel safer. To sum up, autonomous driving has great potential and bright prospects, and the vision in this direction is also great, so it is a direction worth considering.

(3) CV/NLP​

CV and NLP are the two mainstream research directions of traditional deep learning , can be compared to physics and chemistry in basic disciplines, and is the cornerstone of many AI models. CV addresses “what I saw” and NLP addresses “what I heard and said”. If the two directions of CV and NLP are overcome, the machine will learn and understand humans better, and will have many "synaesthesia" capabilities. Therefore, the two directions of CV and NLP are evergreen, especially the explosion of AI painting and Chat-GPT product ideas, which will in turn promote the development of the CV and NLP fields. To sum up, CV and NLP are two directions worth continuing to explore.

(4) AI for Science (AlphaFold2)

The application of AI in the scientific field may attract less attention, but it also has the capabilities Good value and prospects. One of the application directions is quantum computing, which uses the characteristics of the physics community to solve the problem of supercomputing capabilities; the other application direction is protein gene structure prediction (AlphaFold2), which is of great significance to the research and development of new drugs, especially the research and development of cancer-targeted drugs. Most medical technology and the application of AI in the medical and health field were still more theoretical in the past few years. In the past two years, many model results have appeared and have achieved initial results. Therefore, the future vision of AI for Science, especially in the medical field, is still very good. In the next few decades, human life span is likely to be extended due to breakthroughs in AI technology. To sum up, the future value of AI science and even AI medical field is considerable, but this part of the research is still in its preliminary stage, and the period before commercialization is relatively long. Young algorithm engineers can consider trying this kind of cutting-edge competition. Tao, take advantage.

(5) Recommendation system and computational advertising

Recommended system and computational advertising belong to the more traditional "search and advertising" track. Recently, In the past two years, no breakthrough at the methodological level has been achieved. In recent years, conference papers have mostly focused on breakthroughs in some small problems. The search and promotion track relies more on business. If there is no greater breakthrough at the business level, then the algorithm development prospects will be relatively limited. On the other hand, the talent pool in this field is relatively saturated and the competition is fierce, so consider this track carefully.

2. Self Evaluation

In addition to improving the understanding of the Vision level, we must also fully conduct self-evaluation . Self-assessment is mainly considered from 3 dimensions:

(1) What are you good at?

​

What you are best at is often not what you judge by yourself. Generally, you can refer to which part of your past work experience has really impressed others. recognition. The self in other people's impressions may be different from the self in your own eyes, and what you are good at must come from what others recognize.

(2) What are your interests? (Follow your heart)

​

It is very important to find your own points of interest. Everyone has their own period of confusion. During such periods, it is more important to follow your heart and find the areas that truly interest you.

(3) Can it provide you with a decent income?

​

Workplace Life is actually a relatively simple process of matching one's own value with the company's needs. The company exchanges personal value through remuneration. Therefore, a reasonable and decent enough income is very reasonable and very necessary. But don’t just focus on income, work value, growth space and other aspects should also be considered comprehensively.

3. Clear the path and polish the technology

No matter how detailed the planning is, it must ultimately be implemented through action .

1. Path

The first step in action is to clarify your own action path:

(1) Make a short-term plan and a long-term plan

​

① If you only have a short-term plan but lack If you plan for the long term, you will easily become confused after the plan is completed.

② If you only have long-term planning but lack short-term planning, you will easily become a dreamer. If you lack a practical way to implement it, long-term planning will become out of reach.

③ Be sure to combine short-term planning and long-term planning as your own path of action;.

④ The time point defined between short-term planning and long-term planning varies from person to person. It is generally recommended that short-term planning should be based on half a year, and long-term planning should be based on 2-3 years. The year is a cycle.

(2) Look at the growth path from a different perspective​

① As mentioned above Take Career Challenge 2 (from Junior to Senior) as an example. An engineer has been promoted in a large factory (for example, from Alibaba P7 to P9), and the capabilities behind this rank need to be very clear to the engineer. .

② The essence of growth is to go through 4 stages:

a)Troubleshooter - Solve trivial problems: Only by being able to solve trivial problems can you be able to solve bigger problems.

b）Problem Solver - Systematically solve a type of problem: Take the DAU decline of an APP as an example. If the DAU can be reduced Once the problem analysis path is clearly dismantled and corresponding solutions are provided, it will grow from the Troubleshooter stage to the Problem Solver stage.

c）Growth Hacker - Lead the team in the right direction: Go further and clearly analyze and unify the reasons for all DAU decline Once solved, you will have the ability to lead the APP team to achieve DAU growth.

d）Business Pilot - Business Leader: When you grow into a business leader, you will have enough ability and authority to Decide on resources and direction.

③ Clarifying your own growth path is more instructive than clarifying ranks such as P5 or P8.

2. Technology

As an algorithm engineer, excellent technology is a prerequisite:

(1) Maintain technological leadership - (tool, technique, method, way) ​

① Powerful Engineering development ability: As an engineer, development ability is the most basic ability.

② Solid machine learning principles: The principles of machine learning are universal and will lead some analysis ideas. At the same time, machine learning is also a deep learning model and a larger The basis of the model.

③ Tracking of top conferences and cutting-edge directions: Although the directions of top conferences in academia and those in industry may not be synchronized, the cutting-edge directions in academia can often provide Industry provides inspiration for solution ideas.

④ "Best practices" in the industry: Every engineer needs to accumulate "best practices" applicable to specific scenarios in their respective industries. Through accumulation and precipitation, valuable industry experience is gradually formed.

(2) Maintain strong execution ​

All plans, ultimately It all depends on execution, so strong execution and self-driving force are very necessary.

3. Mentality

#In fact, this is true not only in the workplace, but also in many fields: adjust your mentality well, many Things will work out eventually. A good mentality mainly includes the following parts:

(1) Focus Collaborative

Communication and collaboration should focus on solving a few key issues, thereby reducing unnecessary Meaningful meeting.

(2) Take a step forward

As an algorithm engineer, you must To take one more step forward. Learn more about the demands of the product and operations teams, and consider issues from the other side’s perspective. It is necessary to have a "fill-in-the-seat thinking" and work together with other partners such as product and operations to make progress. Each can learn from each other's strengths and weaknesses, and ultimately work together to get things done. Don't reject or even fight the demands of the product or operations teams; don't ignore some of the other party's ideas because the other party isn't experienced enough or thoughtful enough, thereby missing an opportunity to launch a product.

(3) Clear your mind ​

When faced with difficult problems, You might as well try to clear your mind, abandon your inherent thinking patterns, and rethink the core of your business. After removing the inherent burden, you may feel that the current problem may not actually be that complicated and can be solved step by step.

(4) Maintain a relatively strong heart​

Common in the Internet industry The pressure is high and the competition in the industry is very cruel, so you must maintain a strong heart, face difficulties and challenges calmly, and do not be disturbed by the outside world

4. Balance important relationships

(1) Family parent-child relationship

#Family relationship is very important. Very important thing.

(2) Work-Life Balance​

Work is just a part of life , so we still need to better balance work and life, work efficiently and live attentively.

(3) Personal interests​

It is recommended that everyone can develop one or two things Your own interests and hobbies allow your brain to operate in a different way, which is very helpful for relaxing your state.

##4. Summary

This article focuses on 3 parts:

① Challenges faced in three important stages of the workplace.

② How to plan: Vision, self-assessment, and take action.

③ Clear the path, polish the technology, have a positive attitude, and handle important relationships well.

5. Question and Answer Session

Q1: If a position faces a problem such as "35-year-old optimization", is it necessary to study deeply?

A1: This question involves the "mentality" issue mentioned above: "Don't be happy with things, don't be sad with yourself." In fact, everyone will face the "35-year-old problem", which is determined by the environment and the market; and we cannot decide the outside world, but we can decide ourselves; therefore, if we work hard to be ourselves, everything will be fine. As for whether it is necessary to study deeply, the key still lies in the direction of your career planning, as well as the direction and degree of your own deep study. If you want to take the development path of an engineer, it is still necessary to continue to work hard in your own field and lay a solid foundation, which will be very helpful in the future; in addition, you must also consider the expansion of business and management directions, as mentioned in the article "T type of talent". To sum up, the decision-making power of "whether 35 years old is optimized" does not lie with you. Instead of being anxious, it is better to continuously improve your personal abilities and make yourself more active and proactive in the workplace.

#Q2: How do you view the impact of the development of large models on algorithm engineers?

A2: On the one hand, large language models do produce fantastic effects using sufficient training data; but on the other hand, the development of large models is inseparable from products. The "packaging" of the product and the "coat" of the product are thrown away. Its essence is still a classic algorithm model, but the parameter magnitude is huge and the training corpus is richer. The main advantage of the large model is that it has more training data and incorporates some reinforcement learning algorithms to achieve the ultimate in every detail of data training. Therefore, for algorithm engineers, there is no need to be too anxious, but to view these large models positively: First, large models have brought a "boost" to the entire AI algorithm industry, that is, they have been widely recognized by the market from the capital level; secondly, The large model points out the direction for algorithm engineers, and only by combining products, business and technology can they find a way out. As for using AI models to automatically write code, it can be seen as an aid to productivity rather than replacing people.

#Q3: What is the biggest difference between ordinary people and experts in the recommended search field?

#A3: Any technology can be divided into four levels: instrument, technique, method, and Tao. The average person may be more at the level of instrument and technique: using fancy models and using various tricks to adjust parameters, and finally achieve a more satisfactory effect; while masters have often experienced these two levels and find that although these two levels can It can solve some practical problems, but it cannot solve some higher-level problems. This involves Tao and Dharma, and involves deeper essential issues. Take recommendations as an example, how to improve user satisfaction: Because the characterization of user satisfaction is relatively subjective, how to break it down into several objective and quantifiable indicators will test the skills of algorithm engineers. To give a specific example: CTR is a commonly used indicator, which can measure user satisfaction to a certain extent; but if you only optimize the single goal of CTR, it may bring a lot of "clickbait"; therefore, you need to use other indicators To balance this issue; and the selection of indicators requires the accumulation of experience on the one hand, and a deep understanding of the business on the other, which involves the level of law and Tao. Therefore, we must look at "levels" rationally and return to the essence: as algorithm engineers, what we have to do is to use engineering capabilities to solve actual problems and bring value, rather than "show off skills" and play with models; being able to solve problems efficiently is the greatest value to the company.

Q4: If you were to take on an apprentice, which devilish details would you teach him to pay more attention to?

#A4: This first depends on the apprentice’s ability level. If the apprentice is a fresh graduate, he or she cannot have too many requirements and needs to proceed step by step; if the apprentice is a veteran in the industry and has already formed his own methodology in the industry, if he can achieve the task objectives, he does not need to do too many additional steps. Require. From the perspective of team management, since everyone has their own shining points and their own shortcomings, it is difficult to make unified requirements; as a team leader, I will pay more attention to employees' thinking methods and problem-solving ideas, etc. If there are imperfections, incompleteness, or even deviations and errors in these aspects, they need to be pointed out and corrected as soon as possible. In addition, when it comes to the standardization of deliverables, such as online model specifications, code structure specifications, and code comment integrity, more stringent requirements will be imposed; other issues generally will not have too stringent requirements, and will not be subject to strict requirements. It won't involve too many "devil details".

Q5: The article mentioned that the search and promotion field has not developed much in recent years, so what are the future development directions?

#A5: It is not that there has been no development in the field of search and advertising in the past two years, but there has been no major breakthrough in the technical field, and there is still a lot of development in the direction of segmentation. of. However, the development in the past two years has been more at the business level, because the direction of this field is mainly supported by business; therefore, if there is no revolutionary "explosion" in the business, it will be difficult to drive major breakthroughs in technology. As for the future development direction, there should be no new and disruptive model frameworks in the general direction; there will still be a lot of room for development in subdivided areas and directions, which mainly depends on the specific industry and industry. For business directions within the industry, you can pay more attention to relevant top conferences in the industry to find answers.

Q6: An engineer has 3 years of experience in a large factory in the recommended direction. Which direction is more appropriate to switch to?

A6: When switching between algorithm tracks, you must first conduct a self-assessment and have an accurate and comprehensive understanding of your own interests and strengths. For example, if you prefer to study models The method is still more like solving business problems, such as whether there is any subdivision that I am good enough and proficient in in 3 years of experience in a large manufacturer; in addition, we must also consider which development directions are naturally competitive, such as the autonomous driving direction mentioned in the article. This is a competitive direction with promising future and is close to being implemented. Generally speaking, since the recommendation algorithm is directly connected to the business, the recommendation algorithm engineer will have a strong business sensitivity, so it is easy to become a good problem solver in various directions.

Q7: At present, most of the algorithm positions are competed by masters and doctoral students from high-level colleges and universities. How can students from ordinary colleges and universities participate in the competition? Do I need to change my position?

A7: Large factories may have higher requirements for colleges and academic qualifications, but after joining the business, they rely more on personal abilities; students from ordinary colleges and universities You can strive for more internship opportunities and accumulate more project experience without worrying too much about your school background.

#Q8: What should I do if I am too busy at work and have no time to follow the paper?

#A8: Select a few interesting top papers and keep following them; at the same time, pay more attention to platforms such as Zhihu, as well as some industry technology forums, there will be relevant bloggers to help Classify the top conference articles and write summaries. You can read more to find the direction that interests you. Time is squeezed out. Spend half an hour to an hour every day to actually study the paper and refine the key technical points, and then summarize and organize it regularly. In the long run, you can achieve good results.

Q9: What do you think of long video recommendations?

#A9: The operating mechanisms of long video platforms and short video platforms are quite different. Long video recommendations focus more on the content of the video, so algorithm engineers are required to reversely think about the pain points of users consuming long videos from a business perspective; in addition, some units are converting long videos into short videos and extracting highlights from long videos. , convert long videos into short videos through "cutting", and then use the idea of ​​​​short videos to make recommendations. From a product perspective, we can use the idea of ​​​​short videos to recommend long videos through the method of "long videos with short videos"; specifically, clips and trailers of long videos are edited to attract users to watch, and then through the product Path to guide users to watch the full version, and ultimately guide users to become members and increase the user's up value. The enlightenment brought by this is that algorithm engineers still need to return to the product and business perspective to achieve breakthroughs, and cannot be limited to the algorithm perspective.

#Q10: Does algorithm require high academic qualifications? Does social recruitment attach great importance to past experience? If you are not interested in the current technical direction, how to transform?

A10: If an industry is sufficiently involved and competitive pressure is high, the issue of academic qualifications cannot be avoided; if school and academic qualifications are not an advantage, you can add projects To "save the country through curves" through experience. In terms of social recruitment, companies do pay more attention to past experience and hope to introduce candidates' past work experience to empower the company's current business; secondly, they also need to examine the candidates' basic qualities, including self-drive, learning ability, and thinking methods. , coding ability, etc. For changing direction in the workplace, you can first consider changing direction within the company to give yourself a trial period and transition period.

#Q11: The recommendation algorithm has entered a bottleneck period. Does it need to delve deeper into the underlying data (such as Hadoop, SQL, etc.)?

A11: From an engineering perspective, Hadoop, SQL, etc. are indeed the underlying architecture of data; but as algorithm engineers, we should seek breakthroughs in product and business directions. , rather than deeply cultivating and optimizing the underlying structure.

Q12: The boss is from a research institute and lacks engineering implementation experience. However, he pays too much attention to details at work and uses the advantages of competing products to deny and suppress employees. How to improve this situation?

#A12: In fact, a good boss should expand outward rather than squeeze inward. If you unfortunately encounter a boss who is accustomed to squeezing inward, you can guide the boss to look outwards; communicate diligently, understand the boss's pain points, and do a good job in upward management. In addition, if your boss is really difficult to get along with and cannot communicate with you, and you are not good at or interested in the job content, and the job cannot bring you satisfactory income or even affects your life, then it is still recommended to find another job.

Q13: Search, advertising and NLP, which direction is better for employment in the future?

#A13: Search and promotion is closer to the business, while NLP is a relatively basic direction. In recent years, the business development of search advertising and promotion has become relatively mature, but the NLP direction is facing a major problem, that is, it is difficult to achieve product implementation. If you can find a good landing direction from a business perspective, you can try NLP; otherwise, it is recommended to search and promote. In short, in the field of search and promotion, there are ready-made business application topics, which only require the use of algorithms to "solve problems"; in contrast, in the field of NLP, the original questions are very simple, but the answers are very complex.

#Q14: Is there a high threshold for switching algorithm directions during work?

#A14: There are thresholds. It depends on your personal foundation and switching direction. For example, if it is switching within SouGuangTian, ​​the difficulty is relatively small; but if it is switching between CV/NLP and SouGuangTian, ​​it is relatively difficult. Therefore, it still costs a lot to switch the direction of the algorithm during work, especially when social recruitment pays more attention to past experience. However, if you have strong overall qualities, excellent learning ability and self-motivation, and the company's business happens to require such talents, there will be many opportunities to switch directions. If you decide you want to switch directions, you need to choose a path, make a plan and keep taking action.

Q15: What is the future of the field of time series forecasting? Are there any popular directions to recommend?

A15: Quantitative trading scenarios are highly correlated with time series predictions; time series technology is not used in many search and recommendation fields, perhaps in user behavior sequence modeling. It is involved; the field of autonomous driving may also involve some time series prediction, such as FSD path planning, multi-frame concatenation and other scenarios; many technologies in the video field have strong correlation with time series; scenarios such as traffic prediction and customer value prediction Strong correlation with time series forecasting.

Q16: An algorithm engineer has been working in the industry for 3 years and still feels that the work is difficult. Therefore, he wants to know whether the direction of algorithms relies more on talent or hard work?

#A16: First of all, it is necessary to clarify whether the key to "difficulty" lies in reading the paper or solving business problems. In the early stages of career, there will inevitably be periods of confusion, frustration and anxiety. These situations vary from person to person, so it is difficult to give a clear answer. It is more about following your heart and finding your own interests and strengths.

#Q17: What is the difference between recommendation algorithms and data mining?

#A17: Data mining is a more basic technology, and recommendation algorithms are higher-level applications; many data mining technologies are used in recommendation algorithms.

Q18: From the perspective of social recruitment, what level of 3 years of work experience in the direction of e-commerce recommendation algorithms should be achieved?

A18: 3 years of experience can generally reach the quasi-senior level, corresponding to Alibaba's P7, and Tencent's 9th level; with 2 years of further development, you can generally reach Ali's P8 or Tencent level 10 level.

The above is the detailed content of Major Internet companies are facing a 'wave of layoffs'. How can algorithm engineers survive the 'cold winter' of their careers?. For more information, please follow other related articles on the PHP Chinese website!