📝 updated arxiv references

docs/paper.bib

@@ -1,18 +1,19 @@
 
-@misc{masana_class-incremental_2022,
-	title = {Class-incremental learning: survey and performance evaluation on image classification},
-	shorttitle = {Class-incremental learning},
-	url = {http://arxiv.org/abs/2010.15277},
-	doi = {10.48550/arXiv.2010.15277},
-	abstract = {For future learning systems, incremental learning is desirable because it allows for: efficient resource usage by eliminating the need to retrain from scratch at the arrival of new data; reduced memory usage by preventing or limiting the amount of data required to be stored -- also important when privacy limitations are imposed; and learning that more closely resembles human learning. The main challenge for incremental learning is catastrophic forgetting, which refers to the precipitous drop in performance on previously learned tasks after learning a new one. Incremental learning of deep neural networks has seen explosive growth in recent years. Initial work focused on task-incremental learning, where a task-ID is provided at inference time. Recently, we have seen a shift towards class-incremental learning where the learner must discriminate at inference time between all classes seen in previous tasks without recourse to a task-ID. In this paper, we provide a complete survey of existing class-incremental learning methods for image classification, and in particular, we perform an extensive experimental evaluation on thirteen class-incremental methods. We consider several new experimental scenarios, including a comparison of class-incremental methods on multiple large-scale image classification datasets, an investigation into small and large domain shifts, and a comparison of various network architectures.},
-	urldate = {2023-05-09},
-	publisher = {arXiv},
-	author = {Masana, Marc and Liu, Xialei and Twardowski, Bartlomiej and Menta, Mikel and Bagdanov, Andrew D. and van de Weijer, Joost},
-	month = oct,
-	year = {2022},
-	note = {arXiv:2010.15277 [cs]},
-	keywords = {Computer Science - Machine Learning, Computer Science - Computer Vision and Pattern Recognition},
-	file = {arXiv Fulltext PDF:C\:\\Users\\Hundgeburth\\Zotero\\storage\\WZJFLLNS\\Masana et al. - 2022 - Class-incremental learning survey and performance.pdf:application/pdf;arXiv.org Snapshot:C\:\\Users\\Hundgeburth\\Zotero\\storage\\RRJFHSWE\\2010.html:text/html},
+@article{masana_class-incremental_2022,
+	title = {Class-{{Incremental Learning}}: {{Survey}} and {{Performance Evaluation}} on {{Image Classification}}},
+	shorttitle = {Class-{{Incremental Learning}}},
+	author = {Masana, Marc and Liu, Xialei and Twardowski, Bartłomiej and Menta, Mikel and Bagdanov, Andrew D. and family=Weijer, given=Joost, prefix=van de, useprefix=true},
+	date = {2023-05},
+	journaltitle = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
+	volume = {45},
+	number = {5},
+	pages = {5513--5533},
+	issn = {1939-3539},
+	doi = {10.1109/TPAMI.2022.3213473},
+	url = {https://ieeexplore.ieee.org/document/9915459},
+	urldate = {2024-10-08},
+	abstract = {For future learning systems, incremental learning is desirable because it allows for: efficient resource usage by eliminating the need to retrain from scratch at the arrival of new data; reduced memory usage by preventing or limiting the amount of data required to be stored – also important when privacy limitations are imposed; and learning that more closely resembles human learning. The main challenge for incremental learning is catastrophic forgetting, which refers to the precipitous drop in performance on previously learned tasks after learning a new one. Incremental learning of deep neural networks has seen explosive growth in recent years. Initial work focused on task-incremental learning, where a task-ID is provided at inference time. Recently, we have seen a shift towards class-incremental learning where the learner must discriminate at inference time between all classes seen in previous tasks without recourse to a task-ID. In this paper, we provide a complete survey of existing class-incremental learning methods for image classification, and in particular, we perform an extensive experimental evaluation on thirteen class-incremental methods. We consider several new experimental scenarios, including a comparison of class-incremental methods on multiple large-scale image classification datasets, an investigation into small and large domain shifts, and a comparison of various network architectures.},
+	eventtitle = {{{IEEE Transactions}} on {{Pattern Analysis}} and {{Machine Intelligence}}},
 }
 
 @article{lu_learning_2018,
@@ -29,37 +30,34 @@ @article{lu_learning_2018
 	note = {arXiv:2004.05785 [cs, stat]},
 	keywords = {Computer Science - Machine Learning, Statistics - Machine Learning},
 	pages = {1--1},
-	file = {arXiv Fulltext PDF:C\:\\Users\\Hundgeburth\\Zotero\\storage\\JJZ6QZMM\\Lu et al. - 2018 - Learning under Concept Drift A Review.pdf:application/pdf;arXiv.org Snapshot:C\:\\Users\\Hundgeburth\\Zotero\\storage\\DUB9HSNW\\2004.html:text/html},
 }
 
-@misc{hess_procedural_2021,
-	title = {A {Procedural} {World} {Generation} {Framework} for {Systematic} {Evaluation} of {Continual} {Learning}},
-	url = {http://arxiv.org/abs/2106.02585},
-	doi = {10.48550/arXiv.2106.02585},
-	abstract = {Several families of continual learning techniques have been proposed to alleviate catastrophic interference in deep neural network training on non-stationary data. However, a comprehensive comparison and analysis of limitations remains largely open due to the inaccessibility to suitable datasets. Empirical examination not only varies immensely between individual works, it further currently relies on contrived composition of benchmarks through subdivision and concatenation of various prevalent static vision datasets. In this work, our goal is to bridge this gap by introducing a computer graphics simulation framework that repeatedly renders only upcoming urban scene fragments in an endless real-time procedural world generation process. At its core lies a modular parametric generative model with adaptable generative factors. The latter can be used to flexibly compose data streams, which significantly facilitates a detailed analysis and allows for effortless investigation of various continual learning schemes.},
-	urldate = {2023-08-16},
-	publisher = {arXiv},
+@inproceedings{hess_procedural_2021,
+	title = {A {{Procedural World Generation Framework}} for {{Systematic Evaluation}} of {{Continual Learning}}},
 	author = {Hess, Timm and Mundt, Martin and Pliushch, Iuliia and Ramesh, Visvanathan},
-	month = dec,
-	year = {2021},
-	note = {arXiv:2106.02585 [cs]},
-	keywords = {Computer Science - Machine Learning, Computer Science - Computer Vision and Pattern Recognition},
-	file = {arXiv Fulltext PDF:C\:\\Users\\Hundgeburth\\Zotero\\storage\\KV6H2G9R\\Hess et al. - 2021 - A Procedural World Generation Framework for System.pdf:application/pdf;arXiv.org Snapshot:C\:\\Users\\Hundgeburth\\Zotero\\storage\\MWZ4BV8K\\2106.html:text/html},
+	date = {2021-06-08},
+	url = {https://openreview.net/forum?id=LlCQWh8-pwK},
+	urldate = {2024-10-08},
+	abstract = {Several families of continual learning techniques have been proposed to alleviate catastrophic interference in deep neural network training on non-stationary data. However, a comprehensive comparison and analysis of limitations remains largely open due to the inaccessibility to suitable datasets. Empirical examination not only varies immensely between individual works, it further currently relies on contrived composition of benchmarks through subdivision and concatenation of various prevalent static vision datasets. In this work, our goal is to bridge this gap by introducing a computer graphics simulation framework that repeatedly renders only upcoming urban scene fragments in an endless real-time procedural world generation process. At its core lies a modular parametric generative model with adaptable generative factors. The latter can be used to flexibly compose data streams, which significantly facilitates a detailed analysis and allows for effortless investigation of various continual learning schemes.},
+	eventtitle = {Thirty-Fifth {{Conference}} on {{Neural Information Processing Systems Datasets}} and {{Benchmarks Track}} ({{Round}} 1)},
+	langid = {english},
 }
 
-@misc{cossu_is_2021,
-	title = {Is {Class}-{Incremental} {Enough} for {Continual} {Learning}?},
-	url = {http://arxiv.org/abs/2112.02925},
-	doi = {10.3389/frai.2022.829842},
-	abstract = {The ability of a model to learn continually can be empirically assessed in different continual learning scenarios. Each scenario defines the constraints and the opportunities of the learning environment. Here, we challenge the current trend in the continual learning literature to experiment mainly on class-incremental scenarios, where classes present in one experience are never revisited. We posit that an excessive focus on this setting may be limiting for future research on continual learning, since class-incremental scenarios artificially exacerbate catastrophic forgetting, at the expense of other important objectives like forward transfer and computational efficiency. In many real-world environments, in fact, repetition of previously encountered concepts occurs naturally and contributes to softening the disruption of previous knowledge. We advocate for a more in-depth study of alternative continual learning scenarios, in which repetition is integrated by design in the stream of incoming information. Starting from already existing proposals, we describe the advantages such class-incremental with repetition scenarios could offer for a more comprehensive assessment of continual learning models.},
-	urldate = {2023-08-29},
-	publisher = {arXiv},
+@article{cossu_is_2021,
+	title = {Is {{Class-Incremental Enough}} for {{Continual Learning}}?},
 	author = {Cossu, Andrea and Graffieti, Gabriele and Pellegrini, Lorenzo and Maltoni, Davide and Bacciu, Davide and Carta, Antonio and Lomonaco, Vincenzo},
-	month = dec,
-	year = {2021},
-	note = {arXiv:2112.02925 [cs]},
-	keywords = {Computer Science - Artificial Intelligence, Computer Science - Machine Learning},
-	file = {arXiv Fulltext PDF:C\:\\Users\\Hundgeburth\\Zotero\\storage\\DUDSCHWD\\Cossu et al. - 2021 - Is Class-Incremental Enough for Continual Learning.pdf:application/pdf;arXiv.org Snapshot:C\:\\Users\\Hundgeburth\\Zotero\\storage\\9L6DRRXJ\\2112.html:text/html},
+	date = {2022-03-24},
+	journaltitle = {Frontiers in Artificial Intelligence},
+	shortjournal = {Front. Artif. Intell.},
+	volume = {5},
+	publisher = {Frontiers},
+	issn = {2624-8212},
+	doi = {10.3389/frai.2022.829842},
+	url = {https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2022.829842/full},
+	urldate = {2024-10-08},
+	abstract = {{$<$}p{$>$}The ability of a model to learn continually can be empirically assessed in different continual learning scenarios. Each scenario defines the constraints and the opportunities of the learning environment. Here, we challenge the current trend in the continual learning literature to experiment mainly on class-incremental scenarios, where classes present in one experience are never revisited. We posit that an excessive focus on this setting may be limiting for future research on continual learning, since class-incremental scenarios artificially exacerbate catastrophic forgetting, at the expense of other important objectives like forward transfer and computational efficiency. In many real-world environments, in fact, repetition of previously encountered concepts occurs naturally and contributes to softening the disruption of previous knowledge. We advocate for a more in-depth study of alternative continual learning scenarios, in which repetition is integrated by design in the stream of incoming information. Starting from already existing proposals, we describe the advantages such {$<$}italic{$>$}class-incremental with repetition{$<$}/italic{$>$} scenarios could offer for a more comprehensive assessment of continual learning models.{$<$}/p{$>$}},
+	langid = {english},
+	keywords = {catastrophic-forgetting,class-incremental,class-incremental-with-repetition,continual-learning,Lifelong-learning},
 }
 
 @article{wu_wafer_2015,
@@ -133,18 +131,14 @@ @misc{douillard_continuum_2021
 }
 
 @misc{hendrycks_benchmarking_2019,
-	title = {Benchmarking {Neural} {Network} {Robustness} to {Common} {Corruptions} and {Perturbations}},
-	url = {http://arxiv.org/abs/1903.12261},
-	doi = {10.48550/arXiv.1903.12261},
-	abstract = {In this paper we establish rigorous benchmarks for image classifier robustness. Our first benchmark, ImageNet-C, standardizes and expands the corruption robustness topic, while showing which classifiers are preferable in safety-critical applications. Then we propose a new dataset called ImageNet-P which enables researchers to benchmark a classifier's robustness to common perturbations. Unlike recent robustness research, this benchmark evaluates performance on common corruptions and perturbations not worst-case adversarial perturbations. We find that there are negligible changes in relative corruption robustness from AlexNet classifiers to ResNet classifiers. Afterward we discover ways to enhance corruption and perturbation robustness. We even find that a bypassed adversarial defense provides substantial common perturbation robustness. Together our benchmarks may aid future work toward networks that robustly generalize.},
-	urldate = {2024-03-18},
-	publisher = {arXiv},
+	title = {Benchmarking {{Neural Network Robustness}} to {{Common Corruptions}} and {{Perturbations}}},
 	author = {Hendrycks, Dan and Dietterich, Thomas},
-	month = mar,
-	year = {2019},
-	note = {arXiv:1903.12261 [cs, stat]},
-	keywords = {Computer Science - Machine Learning, Statistics - Machine Learning, Computer Science - Computer Vision and Pattern Recognition},
-	file = {arXiv Fulltext PDF:C\:\\Users\\Hundgeburth\\Zotero\\storage\\MFYCHC66\\Hendrycks and Dietterich - 2019 - Benchmarking Neural Network Robustness to Common C.pdf:application/pdf;arXiv.org Snapshot:C\:\\Users\\Hundgeburth\\Zotero\\storage\\KDVQG9FY\\1903.html:text/html},
+	date = {2018-09-27},
+	url = {https://openreview.net/forum?id=HJz6tiCqYm},
+	urldate = {2024-10-08},
+	abstract = {In this paper we establish rigorous benchmarks for image classifier robustness. Our first benchmark, ImageNet-C, standardizes and expands the corruption robustness topic, while showing which classifiers are preferable in safety-critical applications. Then we propose a new dataset called ImageNet-P which enables researchers to benchmark a classifier's robustness to common perturbations. Unlike recent robustness research, this benchmark evaluates performance on common corruptions and perturbations not worst-case adversarial perturbations. We find that there are negligible changes in relative corruption robustness from AlexNet classifiers to ResNet classifiers. Afterward we discover ways to enhance corruption and perturbation robustness. We even find that a bypassed adversarial defense provides substantial common perturbation robustness. Together our benchmarks may aid future work toward networks that robustly generalize.},
+	eventtitle = {International {{Conference}} on {{Learning Representations}}},
+	langid = {english},
 }
 
 @article{paszke_pytorch_nodate,

docs/paper.md

@@ -89,10 +89,8 @@ More declarative ways to build schedules and distributions with special characte
 
 This work was funded by the Austrian Research Promotion Agency (FFG, Project No. 905107).
 
-Special thanks to Benjamin Steinwender, Marius Birkenbach and Nikolaus Neugebauer for their valuable feedback.
-
-I  also want to thank Infineon and KAI for letting me publish this project under a permissive and open license.
-
+Special thanks to Benjamin Steinwender, Marius Birkenbach, Nikolaus Neugebauer, Matthew Feickert, Hoang Anh Ngo and Iztok Fister Jr. for their valuable feedback.
+I also want to thank Infineon and KAI for letting me publish this project under a permissive and open license.
 Finally, I want to thank my university supervisors Thomas Pock and Marc Masana for their guidance.
 
 # References