登录查看更多内容

How do you use online algorithms to make decisions with incomplete data?

由人工智能和领英社区提供技术支持

How do you use online algorithms to make decisions with incomplete data? This is a common challenge in many domains, such as streaming, machine learning, and optimization. Online algorithms are algorithms that process data as it arrives, without knowing the future or the size of the input. They have to make decisions on the fly, based on the information they have seen so far. In this article, we will explore some key concepts and examples of online algorithms, and how they can help you solve problems with incomplete data.

此文章中的业界达人

由社区从 3 条内容中精选。了解更多

Dalmas Chituyi

Building and automating data pipelines to aid ML for supply chain agility and customer intimacy.

1 What are online algorithms?

Online algorithms are algorithms that operate on a sequence of data items, one at a time, without knowing the whole input in advance. They have to make decisions based on the current item and the previous ones, without being able to revise them later. Online algorithms are often compared to offline algorithms, which have access to the entire input before processing it. Online algorithms are useful when the input is too large, too dynamic, or too unpredictable to store or process offline.

添加您的观点

Dalmas Chituyi

Building and automating data pipelines to aid ML for supply chain agility and customer intimacy.
(已编辑)
举报内容
I always think about SGD whenever i came across Online or Incremental learning concepts being discussed. Quite frankly the industry is still shy to exploit this algorithms because they are just to risky in production. However, i think the trick is to combine Batch (have a base model that can regulate model outputs to close to the expected) and Online (Learning from new streams but considering the past outputs when calculating the final output) -doing potentially reduces the competitive ratio and put a human in the loop to monitor and penalize the model in any case.

已翻译

赞

2 Why are online algorithms important?

Online algorithms are important because they can handle situations where offline algorithms cannot. For example, online algorithms can deal with streaming data, such as video, audio, or social media, where the input is continuously generated and needs to be processed in real time. Online algorithms can also handle adaptive learning, where the algorithm has to adjust its behavior based on the feedback it receives from the environment. Online algorithms can also handle optimization problems, where the algorithm has to find the best solution among a changing set of options.

添加您的观点

Dalmas Chituyi

Building and automating data pipelines to aid ML for supply chain agility and customer intimacy.
举报内容
Actually do not quite agree with this observation, in practice i have found that most optimization algorithms thrive from recall, this helps them make the best determination on out of sample instances. There is no evidence that i have come across that suggest Online algorithms make best Optimizers. Yes, they can perform those tasks but due to their nature of "catastrophically forgetting" they can be unreliable.

已翻译

赞

3 How are online algorithms evaluated?

Online algorithms are evaluated based on their performance and their quality. Performance measures how fast and how efficient the algorithm is in processing the input. Quality measures how good the decisions made by the algorithm are, compared to the optimal decisions that could be made if the whole input was known. One way to measure quality is to use the competitive ratio, which is the ratio between the cost of the online algorithm and the cost of the optimal offline algorithm. The lower the competitive ratio, the better the online algorithm.

添加您的观点

Dalmas Chituyi

Building and automating data pipelines to aid ML for supply chain agility and customer intimacy.
(已编辑)
举报内容
I think F1 will be the best Metric to score a Online model. You just want to understand how it pits against the validated data. But most especially just how much' can it recall???

已翻译

赞

4 What are some examples of online algorithms?

Online algorithms are used to solve various problems. For example, the online paging problem requires an algorithm to manage a cache of limited size and minimize the number of cache misses. The least recently used (LRU) algorithm is one online solution that evicts the item that has been used the least recently. For the online matching problem, the goal is to maximize the total value of the matching by using the greedy algorithm, which matches each agent with the highest valued task available at that moment. Finally, for online learning, the goal is to minimize the cumulative error or regret by using the stochastic gradient descent (SGD) algorithm, which updates model or policy parameters in the direction of the gradient of the error or regret.

添加您的观点

5 How can you design online algorithms?

Designing online algorithms can be a difficult task, as one must consider the trade-off between performance and quality, while also dealing with the variability and uncertainty of input. To make this process easier, there are some general principles that can be applied. Heuristics are rules of thumb or simple strategies that can guide the decisions of the online algorithm. Randomness can introduce variability and flexibility into the algorithm, and feedback can provide information or correction to improve its learning or adaptation. Ultimately, online algorithms are powerful tools that can handle streaming, learning, and optimization problems; they can be evaluated by their performance and quality. With heuristics, randomness, and feedback in mind, one can design online algorithms to solve real-world problems with incomplete data, such as caching, matching, or recommending.

添加您的观点

6 Here’s what else to consider

This is a space to share examples, stories, or insights that don’t fit into any of the previous sections. What else would you like to add?

添加您的观点

Algorithms

+ 关注

给文章评分

我们借助人工智能创建了此文章。您认为这篇文章怎么样？

很棒不太好

举报此文章

查看全部

How do you use online algorithms to make decisions with incomplete data?

1

2

3

4

5

6

1 What are online algorithms?

2 Why are online algorithms important?

3 How are online algorithms evaluated?

4 What are some examples of online algorithms?

5 How can you design online algorithms?

6 Here’s what else to consider

Algorithms

给文章评分

感谢您的反馈

更多Algorithms相关文章

更多相关阅读内容

How do you use online algorithms to make decisions with incomplete data?

1

2

3

4

5

6

1 What are online algorithms?

2 Why are online algorithms important?

3 How are online algorithms evaluated?

4 What are some examples of online algorithms?

5 How can you design online algorithms?

6 Here’s what else to consider

Algorithms

给文章评分

感谢您的反馈

查看其他技能