---
title: "SSA: Improving Performance With a Better Scoring Function (2026)"
description: "To counter this, they propose Scaled Signed Averaging (SSA), a trainable, parametric replacement for… 32 references, cited by 1 paper here."
authors: ["Omar Naim", "Swarnadeep Bhar", "Jerome Bolte", "Nicholas Asher"]
year: 2026
venue: "Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)"
source: "https://aclanthology.org/2026.acl-long.1385/"
doi: "10.18653/v1/2026.acl-long.1385"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/naim-2026-ssa-improving-performance-better-scoring/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/naim-2026-ssa-improving-performance-better-scoring/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/17/) / SSA: Improving Performance With a Better Scoring Function

# SSA: Improving Performance With a Better Scoring Function

**Authors:** [Omar Naim](https://md.research.elamur.ai/free/en/authors/omar-naim/), Swarnadeep Bhar, Jerome Bolte and [Nicholas Asher](https://md.research.elamur.ai/free/en/authors/nicholas-asher/) · **Year:** 2026 · **Venue:** [Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)](https://md.research.elamur.ai/free/en/venues/acl/) ([ACL 2026](https://md.research.elamur.ai/free/en/venues/acl/2026/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.acl-long.1385/) · [PDF](https://aclanthology.org/2026.acl-long.1385.pdf) · [DOI 10.18653/v1/2026.acl-long.1385](https://doi.org/10.18653/v1/2026.acl-long.1385) · [OpenAlex](https://openalex.org/W4415314101)

**Citations:** 0 citations (OpenAlex, 2026-10-14) · 1 in our library, of the 149,439 papers whose bibliographies we read

## Summary

Transformers trained for in-context learning (ICL) often break down under simple distribution shifts. The authors trace part of this failure to Softmax, the attention scoring function. They train small GPT-2-style models from scratch on two synthetic ICL tasks. One task judges "every" and "some" statements over number sequences, and the other infers affine functions from examples. In both, a single out-of-range value can dominate attention, and the model then effectively ignores the rest of the context. To counter this, they propose Scaled Signed Averaging (SSA), a trainable, parametric replacement for Softmax. SSA markedly reduces errors on the synthetic tasks, while existing alternatives such as Sparsemax, Entmax and temperature scaling do not give consistent gains. The improvements also carry over to language modelling. A 114M Nemotron-style decoder trained on FineWeb gets lower perplexity and better zero-shot benchmark scores. BabyBERTa encoders trained on AO-CHILDES score higher on many grammatical probes. SSA is therefore presented as a simple drop-in option for both decoder and encoder architectures.

## Key points

- **Controlled ICL setup:** The main model is "12L8AH" (12 layers, 8 heads, 22.5M parameters), trained for 500k steps. Models work well in-distribution but fail on "deviant" inputs, and in the linear task they fail when the function coefficients come from N(0, σ) with σ > 2.
- **Decoder results:** The decoder has 114M parameters, 12 layers, 24 heads and hidden size 768. It was trained on 10B FineWeb tokens for 22k steps. With SSA, perplexity falls from 21.86 to 19.73 on FineWeb and from 24.58 to 22.07 on out-of-distribution Wikipedia text.
- **Zero-shot benchmarks:** SSA scores higher on every reported task. Examples include arc_easy (0.29 → 0.54), boolq (0.38 → 0.56) and hellaswag (0.26 → 0.33).
- **BabyBERTa probes (Huebner et al., 2021):** SSA with exponent 2 helps most on syntactic probes, for example subject–verb agreement across relative clauses rose from 55.5 to 61.55. SSA with exponent 1.5 helps on lexical and morphological probes, for example irregular verbs rose from 70.05 to 78.3. Gains are not uniform, since SSA-2 scores lower than Softmax on a few probes.
- **Limitations:** The authors could not scale beyond 114M parameters. SSA still struggles when both the inputs and the function distribution shift strongly.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

## Institutions

- [Centre National de la Recherche Scientifique](https://md.research.elamur.ai/free/en/labs/centre-national-de-la-recherche-scientifique/) — Research institute, Paris, France · 168 works · 62,369 citations
- [Institut de Recherche en Informatique de Toulouse](https://md.research.elamur.ai/free/en/labs/institut-de-recherche-en-informatique-de-toulouse/) — Research institute, Toulouse, France · 35 works · 1,872 citations
- [Université de Toulouse](https://md.research.elamur.ai/free/en/universities/universite-de-toulouse/) — University, Toulouse, France · 16 works · 1,999 citations
- [Toulouse School of Economics](https://md.research.elamur.ai/free/en/universities/toulouse-school-of-economics/) — University, Toulouse, France · 2 works · 1 citation

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works

## Cite

Cite as: Omar Naim, Swarnadeep Bhar, Jerome Bolte and Nicholas Asher. 2026. SSA: Improving Performance With a Better Scoring Function. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{naim2026ssa,
  title = {{SSA: Improving Performance With a Better Scoring Function}},
  author = {Omar Naim, Swarnadeep Bhar, Jerome Bolte and Nicholas Asher},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-long.1385},
  url = {https://doi.org/10.18653/v1/2026.acl-long.1385}
}
```

- [← previous SRCB at SemEval-2026 Task 5 A Multi-Target Finetuning Framework for Large Language Models with Joint Regression and Text Generation](https://md.research.elamur.ai/free/en/papers/zhang-2026-srcb-semeval-2026-task-5/)
- [next → STAMP: Selective Task-Aware Mechanism for Text Privacy](https://md.research.elamur.ai/free/en/papers/tian-2026-stamp-selective-task-aware-mechanism/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/17/)

## References

32 references, 10 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Nicholas Asher. 2011. Lexical Meaning in Context: A web of words. Cambridge University Press. [find](https://scholar.google.com/scholar?q=Nicholas+Asher%2E+2011%2E+Lexical+Meaning+in+Context%3A+A+web+of+words%2E+Cambridge+University+Press%2E)
2. Nicholas Asher, Swarnadeep Bhar, Akshay Chaturvedi, Julie Hunter, and Soumya Paul. 2023. Limits for learning with large language models. In 12th Joint Conference on Lexical and Computational Semantics (*Sem). Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Nicholas+Asher%2C+Swarnadeep+Bhar%2C+Akshay+Chaturvedi%2C+Julie+Hunter%2C+and+Soumya+Paul%2E+2023%2E+Limits+for+learning+with+large+language+models%2E+In+12th+Joint+Conference+on+Lexical+and+Computational+Semantics)
3. Satwik Bhattamishra, Arkil Patel, Phil Blunsom, and Varun Kanade. 2023. Understanding in-context learning in transformers and llms by learning to learn discrete functions. arXiv preprint arXiv:2310.03016. — [arXiv:2310.03016](https://arxiv.org/abs/2310.03016) [find](https://scholar.google.com/scholar?q=Satwik+Bhattamishra%2C+Arkil+Patel%2C+Phil+Blunsom%2C+and+Varun+Kanade%2E+2023%2E+Understanding+in%2Dcontext+learning+in+transformers+and+llms+by+learning+to+learn+discrete+functions%2E+arXiv+preprint+arXiv%3A2310%2E03)
4. Stella Biderman, Hailey Schoelkopf, Lintang Sutawika, Leo Gao, Jonathan Tow, Baber Abbasi, Alham Fikri Aji, Pawan Sasanka Ammanamanchi, Sidney Black, Jordan Clive, and 1 others. 2024. Lessons from the trenches on reproducible evaluation of language models. arXiv preprint arXiv:2405.14782. [find](https://scholar.google.com/scholar?q=Stella+Biderman%2C+Hailey+Schoelkopf%2C+Lintang+Sutawika%2C+Leo+Gao%2C+Jonathan+Tow%2C+Baber+Abbasi%2C+Alham+Fikri+Aji%2C+Pawan+Sasanka+Ammanamanchi%2C+Sidney+Black%2C+Jordan+Clive%2C+and+1+others%2E+2024%2E+Lessons+from+the)
5. Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, and 1 others. 2020. [Language models are few-shot learners](https://md.research.elamur.ai/free/en/papers/brown-2020-language-models-few-shot-learners/). Advances in neural information processing systems, 33:1877–1901.
6. Akshay Chaturvedi, Swarnadeep Bhar, Soumadeep Saha, Utpal Garain, and Nicholas Asher. 2024. Analyzing Semantic Faithfulness of Language Models via Input Intervention on Question Answering. Computational Linguistics, pages 1–37. [find](https://scholar.google.com/scholar?q=Akshay+Chaturvedi%2C+Swarnadeep+Bhar%2C+Soumadeep+Saha%2C+Utpal+Garain%2C+and+Nicholas+Asher%2E+2024%2E+Analyzing+Semantic+Faithfulness+of+Language+Models+via+Input+Intervention+on+Question+Answering%2E+Computation)
7. Qingxiu Dong, Lei Li, Damai Dai, Ce Zheng, Jingyuan Ma, Rui Li, Heming Xia, Jingjing Xu, Zhiyong Wu, Baobao Chang, and 1 others. 2024. [A survey on in-context learning](https://md.research.elamur.ai/free/en/papers/dong-2023-survey-context-learning/). In Proceedings of the 2024 conference on empirical methods in natural language processing, pages 1107–1128.
8. Shivam Garg, Dimitris Tsipras, Percy S Liang, and Gregory Valiant. 2022. What can transformers learn in-context? a case study of simple function classes. Advances in Neural Information Processing Systems, 35:30583–30598. [find](https://scholar.google.com/scholar?q=Shivam+Garg%2C+Dimitris+Tsipras%2C+Percy+S+Liang%2C+and+Gregory+Valiant%2E+2022%2E+What+can+transformers+learn+in%2Dcontext%3F+a+case+study+of+simple+function+classes%2E+Advances+in+Neural+Information+Processing+Syst)
9. Angeliki Giannou, Liu Yang, Tianhao Wang, Dimitris Papailiopoulos, and Jason D Lee. 2024. How well can transformers emulate in-context newton’s method? arXiv preprint arXiv:2403.03183. — [arXiv:2403.03183](https://arxiv.org/abs/2403.03183) [find](https://scholar.google.com/scholar?q=Angeliki+Giannou%2C+Liu+Yang%2C+Tianhao+Wang%2C+Dimitris+Papailiopoulos%2C+and+Jason+D+Lee%2E+2024%2E+How+well+can+transformers+emulate+in%2Dcontext+newton%E2%80%99s+method%3F+arXiv+preprint+arXiv%3A2403%2E03183%2E+%E2%80%94+%5BarXiv%3A2403%2E0)
10. Geoffrey Hinton, Oriol Vinyals, and Jeff Dean. 2015. [Distilling the knowledge in a neural network](https://md.research.elamur.ai/free/en/papers/hinton-2015-distilling-knowledge-neural-network/). arXiv preprint arXiv:1503.02531.
11. Phu Mon Htut, Jason Phang, Shikha Bordia, and Samuel R Bowman. 2019. Do attention heads in bert track syntactic dependencies? arXiv preprint arXiv:1911.12246. [find](https://scholar.google.com/scholar?q=Phu+Mon+Htut%2C+Jason+Phang%2C+Shikha+Bordia%2C+and+Samuel+R+Bowman%2E+2019%2E+Do+attention+heads+in+bert+track+syntactic+dependencies%3F+arXiv+preprint+arXiv%3A1911%2E12246%2E)
12. Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Liang Wang, Weizhu Chen, and 1 others. 2022. [Lora: Low-rank adaptation of large language models](https://md.research.elamur.ai/free/en/papers/hu-2021-lora-low-rank-adaptation-large/). Iclr, 1(2):3.
13. Philip A Huebner, Elior Sulem, Fisher Cynthia, and Dan Roth. 2021. [Babyberta: Learning more grammar with small-scale child-directed language](https://md.research.elamur.ai/free/en/papers/huebner-2021-babyberta-learning-more-grammar-small/). In Proceedings of the 25th conference on computational natural language learning, pages 624–646.
14. Diederik P Kingma and Jimmy Ba. 2014. [Adam: A method for stochastic optimization](https://md.research.elamur.ai/free/en/papers/kingma-2014-adam-method-stochastic-optimization/). arXiv preprint arXiv:1412.6980.
15. Siddarth Mamidanna, Daking Rai, Ziyu Yao, and Yilun Zhou. 2025. [All for one: Llms solve mental math at the last token with information transferred from other tokens](https://md.research.elamur.ai/free/en/papers/mamidanna-2025-all-one-llms-solve-mental/). In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 30735–30748.
16. Andre Martins and Ramon Astudillo. 2016. From softmax to sparsemax: A sparse model of attention and multi-label classification. In International conference on machine learning, pages 1614–1623. PMLR. [find](https://scholar.google.com/scholar?q=Andre+Martins+and+Ramon+Astudillo%2E+2016%2E+From+softmax+to+sparsemax%3A+A+sparse+model+of+attention+and+multi%2Dlabel+classification%2E+In+International+conference+on+machine+learning%2C+pages+1614%E2%80%931623%2E+PMLR%2E)
17. R Thomas McCoy, Shunyu Yao, Dan Friedman, Mathew D Hardy, and Thomas L Griffiths. 2024. Embers of autoregression show how large language models are shaped by the problem they are trained to solve. Proceedings of the National Academy of Sciences, 121(41):e2322420121. [find](https://scholar.google.com/scholar?q=R+Thomas+McCoy%2C+Shunyu+Yao%2C+Dan+Friedman%2C+Mathew+D+Hardy%2C+and+Thomas+L+Griffiths%2E+2024%2E+Embers+of+autoregression+show+how+large+language+models+are+shaped+by+the+problem+they+are+trained+to+solve%2E+Pro)
18. Omar Naim, Jérôme Bolte, and Nicholas Asher. 2025a. Analyzing limits for in-context learning. arXiv preprint arXiv:2502.03503. — [arXiv:2502.03503](https://arxiv.org/abs/2502.03503) [find](https://scholar.google.com/scholar?q=Omar+Naim%2C+J%C3%A9r%C3%B4me+Bolte%2C+and+Nicholas+Asher%2E+2025a%2E+Analyzing+limits+for+in%2Dcontext+learning%2E+arXiv+preprint+arXiv%3A2502%2E03503%2E+%E2%80%94+%5BarXiv%3A2502%2E03503%5D%28https%3A%2F%2Farxiv%2Eorg%2Fabs%2F2502%2E03503%29)
19. Omar Naim, Guilhem Fouilhé, and Nicholas Asher. 2025b. Re-examining learning linear functions in context. In German Conference on Artificial Intelligence (Künstliche Intelligenz), pages 104–117. Springer. [find](https://scholar.google.com/scholar?q=Omar+Naim%2C+Guilhem+Fouilh%C3%A9%2C+and+Nicholas+Asher%2E+2025b%2E+Re%2Dexamining+learning+linear+functions+in+context%2E+In+German+Conference+on+Artificial+Intelligence+%28K%C3%BCnstliche+Intelligenz%29%2C+pages+104%E2%80%93117%2E+Sprin)
20. Catherine Olsson, Nelson Elhage, Neel Nanda, Nicholas Joseph, Nova DasSarma, Tom Henighan, Ben Mann, Amanda Askell, Yuntao Bai, Anna Chen, and 1 others. 2022. [In-context learning and induction heads](https://md.research.elamur.ai/free/en/papers/olsson-2022-context-learning-induction-heads/). arXiv preprint arXiv:2209.11895.
21. Guilherme Penedo, Hynek Kydlíček, Anton Lozhkov, Margaret Mitchell, Colin Raffel, Leandro Von Werra, Thomas Wolf, and 1 others. 2024. [The fineweb datasets: Decanting the web for the finest text data at scale](https://md.research.elamur.ai/free/en/papers/penedo-2024-fineweb-datasets-decanting-web-finest/). Advances in Neural Information Processing Systems, 37:30811–30849.
22. Ben Peters, Vlad Niculae, and André FT Martins. 2019. Sparse sequence-to-sequence models. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 1504–1519. [find](https://scholar.google.com/scholar?q=Ben+Peters%2C+Vlad+Niculae%2C+and+Andr%C3%A9+FT+Martins%2E+2019%2E+Sparse+sequence%2Dto%2Dsequence+models%2E+In+Proceedings+of+the+57th+Annual+Meeting+of+the+Association+for+Computational+Linguistics%2C+pages+1504%E2%80%931519%2E)
23. Zhen Qin, Weixuan Sun, Hui Deng, Dongxu Li, Yunshen Wei, Baohong Lv, Junjie Yan, Lingpeng Kong, and Yiran Zhong. 2022. cosformer: Rethinking softmax in attention. arXiv preprint arXiv:2202.08791. — [arXiv:2202.08791](https://arxiv.org/abs/2202.08791) [find](https://scholar.google.com/scholar?q=Zhen+Qin%2C+Weixuan+Sun%2C+Hui+Deng%2C+Dongxu+Li%2C+Yunshen+Wei%2C+Baohong+Lv%2C+Junjie+Yan%2C+Lingpeng+Kong%2C+and+Yiran+Zhong%2E+2022%2E+cosformer%3A+Rethinking+softmax+in+attention%2E+arXiv+preprint+arXiv%3A2202%2E08791%2E+%E2%80%94+%5Ba)
24. Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, Ilya Sutskever, and 1 others. 2019. Language models are unsupervised multitask learners. OpenAI blog, 1(8):9. [find](https://scholar.google.com/scholar?q=Alec+Radford%2C+Jeffrey+Wu%2C+Rewon+Child%2C+David+Luan%2C+Dario+Amodei%2C+Ilya+Sutskever%2C+and+1+others%2E+2019%2E+Language+models+are+unsupervised+multitask+learners%2E+OpenAI+blog%2C+1%288%29%3A9%2E)
25. Allan Raventós, Mansheej Paul, Feng Chen, and Surya Ganguli. 2023. Pretraining task diversity and the emergence of non-bayesian in-context learning for regression. Advances in neural information processing systems, 36:14228–14246. [find](https://scholar.google.com/scholar?q=Allan+Ravent%C3%B3s%2C+Mansheej+Paul%2C+Feng+Chen%2C+and+Surya+Ganguli%2E+2023%2E+Pretraining+task+diversity+and+the+emergence+of+non%2Dbayesian+in%2Dcontext+learning+for+regression%2E+Advances+in+neural+information+proce)
26. Pradyut Sekhsaria, Marcel Mateos Salles, Hai Huang, and Randall Balestriero. 2025. Lora users beware: A few spurious tokens can manipulate your finetuned model. arXiv preprint arXiv:2506.11402. — [arXiv:2506.11402](https://arxiv.org/abs/2506.11402) [find](https://scholar.google.com/scholar?q=Pradyut+Sekhsaria%2C+Marcel+Mateos+Salles%2C+Hai+Huang%2C+and+Randall+Balestriero%2E+2025%2E+Lora+users+beware%3A+A+few+spurious+tokens+can+manipulate+your+finetuned+model%2E+arXiv+preprint+arXiv%3A2506%2E11402%2E+%E2%80%94+%5BarX)
27. Yuandong Tian, Yiping Wang, Beidi Chen, and Simon S Du. 2023. Scan and snap: Understanding training dynamics and token composition in 1-layer transformer. Advances in neural information processing systems, 36:71911–71947. [find](https://scholar.google.com/scholar?q=Yuandong+Tian%2C+Yiping+Wang%2C+Beidi+Chen%2C+and+Simon+S+Du%2E+2023%2E+Scan+and+snap%3A+Understanding+training+dynamics+and+token+composition+in+1%2Dlayer+transformer%2E+Advances+in+neural+information+processing+sys)
28. Jesse Vig. 2019. Visualizing attention in transformer-based language representation models. arXiv preprint arXiv:1904.02679. — [arXiv:1904.02679](https://arxiv.org/abs/1904.02679) [find](https://scholar.google.com/scholar?q=Jesse+Vig%2E+2019%2E+Visualizing+attention+in+transformer%2Dbased+language+representation+models%2E+arXiv+preprint+arXiv%3A1904%2E02679%2E+%E2%80%94+%5BarXiv%3A1904%2E02679%5D%28https%3A%2F%2Farxiv%2Eorg%2Fabs%2F1904%2E02679%29)
29. Sang Michael Xie, Aditi Raghunathan, Percy Liang, and Tengyu Ma. 2021. [An explanation of in-context learning as implicit bayesian inference](https://md.research.elamur.ai/free/en/papers/xie-2021-explanation-context-learning-implicit-bayesian/). arXiv preprint arXiv:2111.02080.
30. Jiacheng Ye, Jiahui Gao, Shansan Gong, Lin Zheng, Xin Jiang, Zhenguo Li, and Lingpeng Kong. 2024. Beyond autoregression: Discrete diffusion for complex reasoning and planning. arXiv preprint arXiv:2410.14157. — [arXiv:2410.14157](https://arxiv.org/abs/2410.14157) [find](https://scholar.google.com/scholar?q=Jiacheng+Ye%2C+Jiahui+Gao%2C+Shansan+Gong%2C+Lin+Zheng%2C+Xin+Jiang%2C+Zhenguo+Li%2C+and+Lingpeng+Kong%2E+2024%2E+Beyond+autoregression%3A+Discrete+diffusion+for+complex+reasoning+and+planning%2E+arXiv+preprint+arXiv%3A241)
31. Ruiqi Zhang, Spencer Frei, and Peter L Bartlett. 2024. Trained transformers learn linear models in-context. Journal of Machine Learning Research, 25(49):1–55. [find](https://scholar.google.com/scholar?q=Ruiqi+Zhang%2C+Spencer+Frei%2C+and+Peter+L+Bartlett%2E+2024%2E+Trained+transformers+learn+linear+models+in%2Dcontext%2E+Journal+of+Machine+Learning+Research%2C+25%2849%29%3A1%E2%80%9355%2E)
32. Chuanyang Zheng, Yihang Gao, Guoxuan Chen, Han Shi, Jing Xiong, Xiaozhe Ren, Chao Huang, Zhenguo Li, and Yu Li. 2025. Self-adjust softmax. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 7827–7847. [find](https://scholar.google.com/scholar?q=Chuanyang+Zheng%2C+Yihang+Gao%2C+Guoxuan+Chen%2C+Han+Shi%2C+Jing+Xiong%2C+Xiaozhe+Ren%2C+Chao+Huang%2C+Zhenguo+Li%2C+and+Yu+Li%2E+2025%2E+Self%2Dadjust+softmax%2E+In+Proceedings+of+the+2025+Conference+on+Empirical+Methods+in)

*Updated 31 October 2026*