Matches in SemOpenAlex for { <https://semopenalex.org/work/W3199382086> ?p ?o ?g. }
- W3199382086 abstract "The cross entropy loss is widely used due to its effectiveness and solid theoretical grounding. However, as training progresses, the loss tends to focus on hard to classify samples, which may prevent the network from obtaining gains in performance. While most work in the field suggest ways to classify hard negatives, we suggest to strategically leave hard negatives behind, in order to focus on misclassified samples with higher probabilities. We show that adding to the optimization goal the expectation loss, which is a better approximation of the zero-one loss, helps the network to achieve better accuracy. We, therefore, propose to shift between the two losses during training, focusing more on the expectation loss gradually during the later stages of training. Our experiments show that the new training protocol improves performance across a diverse set of classification domains, including computer vision, natural language processing, tabular data, and sequences. Our code and scripts are available at supplementary." @default.
- W3199382086 created "2021-09-27" @default.
- W3199382086 creator A5000106487 @default.
- W3199382086 creator A5052864256 @default.
- W3199382086 creator A5078102229 @default.
- W3199382086 date "2021-09-12" @default.
- W3199382086 modified "2023-09-27" @default.
- W3199382086 title "Mixing between the Cross Entropy and the Expectation Loss Terms." @default.
- W3199382086 cites W103934657 @default.
- W3199382086 cites W1499021337 @default.
- W3199382086 cites W1565746575 @default.
- W3199382086 cites W1941659294 @default.
- W3199382086 cites W1945616565 @default.
- W3199382086 cites W2129160848 @default.
- W3199382086 cites W2142859603 @default.
- W3199382086 cites W2173398862 @default.
- W3199382086 cites W2189962665 @default.
- W3199382086 cites W2194775991 @default.
- W3199382086 cites W2440599146 @default.
- W3199382086 cites W2460937040 @default.
- W3199382086 cites W2549139847 @default.
- W3199382086 cites W2604156156 @default.
- W3199382086 cites W2605488176 @default.
- W3199382086 cites W2614635052 @default.
- W3199382086 cites W2640329709 @default.
- W3199382086 cites W2743473392 @default.
- W3199382086 cites W2747685395 @default.
- W3199382086 cites W2790626470 @default.
- W3199382086 cites W2798869704 @default.
- W3199382086 cites W2801243887 @default.
- W3199382086 cites W2803187616 @default.
- W3199382086 cites W2804386825 @default.
- W3199382086 cites W2883780447 @default.
- W3199382086 cites W2895171208 @default.
- W3199382086 cites W2896457183 @default.
- W3199382086 cites W2899771611 @default.
- W3199382086 cites W2945557596 @default.
- W3199382086 cites W2949427019 @default.
- W3199382086 cites W2950348158 @default.
- W3199382086 cites W2950906520 @default.
- W3199382086 cites W2951959408 @default.
- W3199382086 cites W2952020226 @default.
- W3199382086 cites W2952659618 @default.
- W3199382086 cites W2952927437 @default.
- W3199382086 cites W2962749006 @default.
- W3199382086 cites W2962839956 @default.
- W3199382086 cites W2963310665 @default.
- W3199382086 cites W2963389226 @default.
- W3199382086 cites W2963454111 @default.
- W3199382086 cites W2963857521 @default.
- W3199382086 cites W2963874210 @default.
- W3199382086 cites W2963899139 @default.
- W3199382086 cites W2964072432 @default.
- W3199382086 cites W2964082701 @default.
- W3199382086 cites W2979811391 @default.
- W3199382086 cites W3118608800 @default.
- W3199382086 cites W3120740533 @default.
- W3199382086 cites W607505555 @default.
- W3199382086 hasPublicationYear "2021" @default.
- W3199382086 type Work @default.
- W3199382086 sameAs 3199382086 @default.
- W3199382086 citedByCount "0" @default.
- W3199382086 crossrefType "posted-content" @default.
- W3199382086 hasAuthorship W3199382086A5000106487 @default.
- W3199382086 hasAuthorship W3199382086A5052864256 @default.
- W3199382086 hasAuthorship W3199382086A5078102229 @default.
- W3199382086 hasConcept C106301342 @default.
- W3199382086 hasConcept C111919701 @default.
- W3199382086 hasConcept C11413529 @default.
- W3199382086 hasConcept C119857082 @default.
- W3199382086 hasConcept C120665830 @default.
- W3199382086 hasConcept C121332964 @default.
- W3199382086 hasConcept C124101348 @default.
- W3199382086 hasConcept C153180895 @default.
- W3199382086 hasConcept C154945302 @default.
- W3199382086 hasConcept C167981619 @default.
- W3199382086 hasConcept C177264268 @default.
- W3199382086 hasConcept C192209626 @default.
- W3199382086 hasConcept C199360897 @default.
- W3199382086 hasConcept C41008148 @default.
- W3199382086 hasConcept C43126263 @default.
- W3199382086 hasConcept C51632099 @default.
- W3199382086 hasConcept C61423126 @default.
- W3199382086 hasConcept C62520636 @default.
- W3199382086 hasConceptScore W3199382086C106301342 @default.
- W3199382086 hasConceptScore W3199382086C111919701 @default.
- W3199382086 hasConceptScore W3199382086C11413529 @default.
- W3199382086 hasConceptScore W3199382086C119857082 @default.
- W3199382086 hasConceptScore W3199382086C120665830 @default.
- W3199382086 hasConceptScore W3199382086C121332964 @default.
- W3199382086 hasConceptScore W3199382086C124101348 @default.
- W3199382086 hasConceptScore W3199382086C153180895 @default.
- W3199382086 hasConceptScore W3199382086C154945302 @default.
- W3199382086 hasConceptScore W3199382086C167981619 @default.
- W3199382086 hasConceptScore W3199382086C177264268 @default.
- W3199382086 hasConceptScore W3199382086C192209626 @default.
- W3199382086 hasConceptScore W3199382086C199360897 @default.
- W3199382086 hasConceptScore W3199382086C41008148 @default.
- W3199382086 hasConceptScore W3199382086C43126263 @default.
- W3199382086 hasConceptScore W3199382086C51632099 @default.