REVIEW 5 cited by
Friends and Foes in Learning from Noisy Labels
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
Friends and Foes in Learning from Noisy Labels
read the original abstract
Learning from examples with noisy labels has attracted increasing attention recently. But, this paper will show that the commonly used CIFAR-based datasets and the accuracy evaluation metric used in the literature are both inappropriate in this context. An alternative valid evaluation metric and new datasets are proposed in this paper to promote proper research and evaluation in this area. Then, friends and foes are identified from existing methods as technical components that are either beneficial or detrimental to deep learning from noisy labeled examples, respectively, and this paper improves and combines technical components from the friends category, including self-supervised learning, new warmup strategy, instance filtering and label correction. The resulting F&F method significantly outperforms existing methods on the proposed nCIFAR datasets and the real-world Clothing1M dataset.
Forward citations
Cited by 5 Pith papers
-
Spectropolarimetric detection of baryonic mass loading in a transient relativistic jet: application to the black hole X-ray binary Swift J1727.8$-$1613
Transient Faraday-complex spectropolarimetric structure detected during radio flaring of Swift J1727 implies internal Faraday rotation from electron-proton jet plasma with rotating mass ~10^21 g, a small fraction of a...
-
How Low Can We Go? Minimum Spectroscopic Requirements For Supernova Subtype Classification
ABC-SN classifies ten supernova subtypes with no performance loss down to R_λ=50 and SNR=5, and only minimal loss at R_λ=25.
-
The Environments of Luminous Fast Blue Optical Transients: Evidence for a Compact Object and Wolf-Rayet Star Merger Origin
LFBOT hosts are star-forming and moderately metal-poor, with many events offset from bright light, favoring a compact-object–Wolf-Rayet merger origin over TDEs or standard CCSNe.
-
Tightening Bounds on Warm Dark Matter with High-Redshift Gamma-Ray Bursts
Using the redshift distribution of 118 luminous Swift GRBs, warm dark matter particles are bounded to mx ≥ 1.3 keV at 95% CL (≥3.4 keV if GRBs exactly trace the SFR).
-
Gamma-ray bursts reveal the history and faint contributors of cosmic reionization
Long gamma-ray bursts imply a higher cosmic star formation rate density at z>6 than galaxy surveys detect, sufficient to drive reionization with moderate ionizing efficiency and escape fraction, implying a large popul...
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.