Rethinking the Optimization of Average Precision: Only Penalizing Negative Instances before Positive Ones Is Enough

نویسندگان

چکیده

Optimising the approximation of Average Precision (AP) has been widely studied for image retrieval. Limited by definition AP, such methods consider both negative and positive instances ranking before each instance. However, we claim that only penalizing ones is enough, because loss comes from these instances. To this end, propose a novel loss, namely Penalizing Negative Positive (PNP), which can directly minimize number one. In addition, AP-based adopt fixed sub-optimal gradient assignment strategy. Therefore, systematically investigate different solutions via constructing derivative functions resulting in PNP-I with increasing PNP-D decreasing ones. focuses more on hard assigning larger gradients to them tries make all relevant closer. contrast, pays less attention slowly corrects them. For most real-world data, one class usually contains several local clusters. blindly gathers clusters while keeps as they were. superior. Experiments three standard retrieval datasets show consistent results above analysis. Extensive evaluations demonstrate achieves state-of-the-art performance. Code available at https://github.com/interestingzhuo/PNPloss

برای دانلود باید عضویت طلایی داشته باشید

برای دانلود متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

منابع مشابه

Efficient Optimization for Average Precision SVM

The accuracy of information retrieval systems is often measured using average precision (AP). Given a set of positive (relevant) and negative (non-relevant) samples, the parameters of a retrieval system can be estimated using the AP-SVM framework, which minimizes a regularized convex upper bound on the empirical AP loss. However, the high computational complexity of loss-augmented inference, wh...

متن کامل

Surveillance of hepatocellular carcinoma: is only ultrasound enough?

Early detection of hepatocellular carcinoma (HCC) is very important because the 5-year survival rate of early stage HCC (about 40% to 70%) is better than that of advanced stage HCC (less than 5%). In some countries, high-risk patients undergo a hepatocarcinoma surveillance program using ultrasound and serum alphafetoprotein (AFP) levels. However, there are several drawbacks of surveillance prog...

متن کامل

Rethinking emotion: cognitive reappraisal is an effective positive and negative emotion regulation strategy in bipolar disorder.

Bipolar disorder involves difficulties with emotion regulation, yet the precise nature of these emotion regulatory difficulties is unclear. The current study examined whether individuals with remitted bipolar I disorder (n = 23) and healthy controls (n = 23) differ in their ability to use one effective and common form of emotion regulation, cognitive reappraisal. Positive, negative, and neutral...

متن کامل

How much precision in reporting statistics is enough?

Depending on the accuracy of the tools we employ in our research, each variable is measured within a certain degree of precision. For example, in most clinical studies on adults, age is measured in years. Generally, measuring the age with more accuracy in such studies is neither necessary nor of any particular importance. However, we might measure blood pH in the same study with two or even thr...

متن کامل

ذخیره در منابع من


  با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید

ژورنال

عنوان ژورنال: Proceedings of the ... AAAI Conference on Artificial Intelligence

سال: 2022

ISSN: ['2159-5399', '2374-3468']

DOI: https://doi.org/10.1609/aaai.v36i2.20042