Submission declined on 12 December 2025 by ChrysGalley (talk). This draft is not written from a neutral point of view. Wikipedia articles must be written neutrally in a formal, impersonal, and dispassionate way. They should not read like a blog post, advertisement, or fan page. Rewrite the draft to remove: promotional language: see Words to watch; personal commentary: opinions or direct addresses to the reader; informal language. Instead, only summarize in your own words a range of independent, reliable, published sources that discuss the subject. If you would like to continue working on the submission, click on the "Edit" tab at the top of the window. If you have not resolved the issues listed above, your draft will be declined again and potentially deleted. If you need extra help, please ask us a question at the AfC Help Desk or get live help from experienced editors. Please do not remove reviewer comments or this notice until the submission is accepted. Where to get help If you need help editing or submitting your draft, please ask us a question at the AfC Help Desk or get live help from experienced editors. These venues are only for help with editing and the submission process, not to get reviews. If you need feedback on your draft, or if the review is taking a lot of time, you can try asking for help on the talk page of a relevant WikiProject. Some WikiProjects are more active than others so a speedy reply is not guaranteed. How to improve a draft Wikipedia:Contributing to Wikipedia – a basic overview on how to edit Wikipedia. Help:Wikitext – how to use the markup Help:Referencing for beginners – how to include references Wikipedia:Article development – how to develop your article Wikipedia:Writing better articles – how to improve your article Wikipedia:Verifiability – make sure your article includes reliable third-party sources You can also browse Wikipedia:Featured articles and Wikipedia:Good articles to find examples of Wikipedia's best writing on topics similar to your proposed article. Improving your odds of a speedy review To improve your odds of a faster review, tag your draft with relevant WikiProject tags using the button below. This will let reviewers know a new draft has been submitted in their area of interest. For instance, if you wrote about a female astronomer, you would want to add the Biography, Astronomy, and Women scientists tags. Add tags to your draft Editor resources Easy tools: Citation bot (help) | Advanced: Fix bare URLs Declined by ChrysGalley 5 months ago. Last edited by ChrysGalley 5 months ago. Reviewer: Inform author. Resubmit Please note that if the issues are not fixed, the draft will be declined again.

• Comment: It is a lot easier to understand the sources than the article. The Background section is particularly badly phrased, and the article was possibly constructed with assistance from AI / LLM. The Lead section presumed that someone was either familiar with, or would follow through on, the Tsung-Yi Lin (etc) source. Instead the Lead should permit an educated reader to follow along, assisted by wikilinks if necessary. See WP:EXPLAINLEAD.
  I can see a relatively straightforward rewrite, by a human, could make this article fully understandable, with the possible exception of the Formulation section.
  The sources show a "primary class" error, which looks to be an arXiv category identifier error. I'm not sure how a human would make this error, since it's not a parameter on "cite journal". ChrysGalley (talk) 16:24, 12 December 2025 (UTC)

• Comment: Make sure to WP:Make technical articles understandable. Commandant Quacks-a-lot (talk) 19:34, 10 December 2025 (UTC)

Focal loss is a loss function used in machine learning to address class imbalance in classification tasks, especially in dense object detection problems. It was introduced by Lin et al. (2017) in the RetinaNet architecture.^[1] Currently, focal loss is often a preferred training loss, especially in computer vision tasks such as classification and segmentation.

In classification problems with a significant imbalance between foreground and background classes, standard loss functions such as cross-entropy may be dominated by easy, majority class examples. As a result, the model may learn poorly on hard or minority class examples. Focal loss modifies the standard cross-entropy loss to focus training on hard examples and down-weight the contribution of easy ones.^[1]

Let ${\displaystyle p_{t}}$ be the model's estimated probability for the true class label ${\displaystyle t\in [1:K]}$ where ${\displaystyle K}$ is the number of classes. The focal loss is defined as: ^[1]

${\displaystyle \mathrm {FL} (p_{t})=-\alpha _{t}(1-p_{t})^{\gamma }\log(p_{t})}$

where:

• ${\displaystyle \gamma \geq 0}$ is the focusing parameter that adjusts the rate at which easy examples are down-weighted,
• ${\displaystyle \alpha _{t}\in [0,1]}$ is a weighting factor to address class imbalance.

When ${\displaystyle \gamma =0}$, focal loss reduces to the standard cross-entropy loss. Larger values of ${\displaystyle \gamma }$ place more focus on hard, misclassified examples.

The term ${\displaystyle (1-p_{t})^{\gamma }}$ is referred to as the focusing factor. It down-weights examples that the model already classifies with high confidence (that is, when ${\displaystyle p_{t}\approx 1}$) and preserves the contribution of harder, misclassified examples (when ${\displaystyle p_{t}\ll 1}$). As ${\displaystyle \gamma }$ increases, the loss for well-classified samples is substantially reduced, while the loss for hard samples is affected much less. This focusing effect reshapes the gradient so that training emphasizes difficult examples and mitigates the dominance of abundant "easy" negatives in settings with severe class imbalance.

Focal loss was originally proposed for use in the RetinaNet^[1]? architecture, which achieved state-of-the-art performance in object detection on benchmarks such as COCO. It has since been adopted in a variety of tasks, including:

• Dense object detection^[1]
• Semantic segmentation^[2]
• Medical image analysis ^[3]
• Multi-class classification under imbalance ^[4]

Focal loss has also found application outside of machine learning in areas such as lossy compression.^[5]

Several extensions of focal loss have been proposed to address specific types of class imbalance or to adapt the loss to different prediction tasks.

• Generalized focal loss with tunable curvature for other divergence measures^[6]
• Focal Tversky loss for segmentation tasks^[7]. The focal Tversky loss replaces the cross-entropy base with the Tversky index, making it particularly effective for medical image segmentation where foreground regions are small and highly asymmetric.
• Asymmetric focal loss for multi-label classification^[8]. Asymmetric focal loss introduces different focusing strengths for positive and negative classes, allowing the loss to penalize false negatives more heavily than false positives.

• Cross entropy
• Loss function
• Object detection