home.social

#overfitting — Public Fediverse posts

Live and recent posts from across the Fediverse tagged #overfitting, aggregated by home.social.

fetched live
  1. #Compiler outputs are deterministic, whereas #LLM outputs are probabilistic.

    Right now #LLM is getting the same regurgitated data again & again. Soon #overfitting would make it deterministic.

  2. #Compiler outputs are deterministic, whereas #LLM outputs are probabilistic.

    Right now #LLM is getting the same regurgitated data again & again. Soon #overfitting would make it deterministic.

  3. #Compiler outputs are deterministic, whereas #LLM outputs are probabilistic.

    Right now #LLM is getting the same regurgitated data again & again. Soon #overfitting would make it deterministic.

  4. #Compiler outputs are deterministic, whereas #LLM outputs are probabilistic.

    Right now #LLM is getting the same regurgitated data again & again. Soon #overfitting would make it deterministic.

  5. #Compiler outputs are deterministic, whereas #LLM outputs are probabilistic.

    Right now #LLM is getting the same regurgitated data again & again. Soon #overfitting would make it deterministic.

  6. 🔑 3 cuidados esenciales para entrenar redes neuronales:
    1️⃣ Minimiza el error cuadrático medio – Ajusta los pesos de forma iterativa para acercarte a la realidad.
    2️⃣ Normaliza y escala los datos – Evita que variables con rangos grandes distorsionen el aprendizaje.
    3️⃣ Divide tus datos – Usa un conjunto de validación externo para monitorear la calidad del modelo.
    🧠 Un modelo que solo sirve para entrenar no sirve para nada.
    #MachineLearning #IA #DataScience #RedesNeuronales #Overfitting #DeepLearning

  7. 🔑 3 cuidados esenciales para entrenar redes neuronales:
    1️⃣ Minimiza el error cuadrático medio – Ajusta los pesos de forma iterativa para acercarte a la realidad.
    2️⃣ Normaliza y escala los datos – Evita que variables con rangos grandes distorsionen el aprendizaje.
    3️⃣ Divide tus datos – Usa un conjunto de validación externo para monitorear la calidad del modelo.
    🧠 Un modelo que solo sirve para entrenar no sirve para nada.
    #MachineLearning #IA #DataScience #RedesNeuronales #Overfitting #DeepLearning

  8. I was working this week on a UI to help users understand when they’re #overfitting their data. Coming up with ideas for the #UI/#UX was proving to be difficult.

    My experience 2 months ago with #Claude was terrible. But this time the output from #ClaudeDesign, was pretty good: training vs prediction, noise and model complexity controls, and comparing fits across degrees.

    Needs improvement, but its a good starting point. AI in design is going to be a thing.

    #ux #ui #datascience #prototyping

  9. I was working this week on a UI to help users understand when they’re #overfitting their data. Coming up with ideas for the #UI/#UX was proving to be difficult.

    My experience 2 months ago with #Claude was terrible. But this time the output from #ClaudeDesign, was pretty good: training vs prediction, noise and model complexity controls, and comparing fits across degrees.

    Needs improvement, but its a good starting point. AI in design is going to be a thing.

    #ux #ui #datascience #prototyping

  10. I was working this week on a UI to help users understand when they’re #overfitting their data. Coming up with ideas for the #UI/#UX was proving to be difficult.

    My experience 2 months ago with #Claude was terrible. But this time the output from #ClaudeDesign, was pretty good: training vs prediction, noise and model complexity controls, and comparing fits across degrees.

    Needs improvement, but its a good starting point. AI in design is going to be a thing.

    #ux #ui #datascience #prototyping

  11. I was working this week on a UI to help users understand when they’re #overfitting their data. Coming up with ideas for the #UI/#UX was proving to be difficult.

    My experience 2 months ago with #Claude was terrible. But this time the output from #ClaudeDesign, was pretty good: training vs prediction, noise and model complexity controls, and comparing fits across degrees.

    Needs improvement, but its a good starting point. AI in design is going to be a thing.

    #ux #ui #datascience #prototyping

  12. I was working this week on a UI to help users understand when they’re #overfitting their data. Coming up with ideas for the #UI/#UX was proving to be difficult.

    My experience 2 months ago with #Claude was terrible. But this time the output from #ClaudeDesign, was pretty good: training vs prediction, noise and model complexity controls, and comparing fits across degrees.

    Needs improvement, but its a good starting point. AI in design is going to be a thing.

    #ux #ui #datascience #prototyping

  13. Furthermore, we were discussing overfitting as another major problem with machine learning. SImply memorising the data doesn't help, when you have to make predictions over unknown data. When overfitting, the model looses the ability to generalise...

    #AI #lecture #machine learning #KDAI2026 #overfitting #datascience #data @fiz_karlsruhe @fizise #knowledge

  14. Furthermore, we were discussing overfitting as another major problem with machine learning. SImply memorising the data doesn't help, when you have to make predictions over unknown data. When overfitting, the model looses the ability to generalise...

    #AI #lecture #machine learning #KDAI2026 #overfitting #datascience #data @fiz_karlsruhe @fizise #knowledge

  15. Furthermore, we were discussing overfitting as another major problem with machine learning. SImply memorising the data doesn't help, when you have to make predictions over unknown data. When overfitting, the model looses the ability to generalise...

    #AI #lecture #machine learning #KDAI2026 #overfitting #datascience #data @fiz_karlsruhe @fizise #knowledge

  16. Furthermore, we were discussing overfitting as another major problem with machine learning. SImply memorising the data doesn't help, when you have to make predictions over unknown data. When overfitting, the model looses the ability to generalise...

    #AI #lecture #machine learning #KDAI2026 #overfitting #datascience #data @fiz_karlsruhe @fizise #knowledge

  17. Furthermore, we were discussing overfitting as another major problem with machine learning. SImply memorising the data doesn't help, when you have to make predictions over unknown data. When overfitting, the model looses the ability to generalise...

    #AI #lecture #machine learning #KDAI2026 #overfitting #datascience #data @fiz_karlsruhe @fizise #knowledge

  18. 🔮 Behold, a revolutionary tome unveiling the mystical art of... splitting data sets! 🎩✨ Dive into a world where machine learning geniuses compete in a bizarre contest of who-can-overfit-the-best, and where #benchmarks are the sacred cow 🐄 that everyone loves to hate but won't stop worshipping. Spoiler: it's #groundbreaking, like discovering water is wet. 💧🤯
    mlbenchmarks.org/00-preface.ht #dataScience #machineLearning #overfitting #techHumor #HackerNews #ngated

  19. 🔮 Behold, a revolutionary tome unveiling the mystical art of... splitting data sets! 🎩✨ Dive into a world where machine learning geniuses compete in a bizarre contest of who-can-overfit-the-best, and where #benchmarks are the sacred cow 🐄 that everyone loves to hate but won't stop worshipping. Spoiler: it's #groundbreaking, like discovering water is wet. 💧🤯
    mlbenchmarks.org/00-preface.ht #dataScience #machineLearning #overfitting #techHumor #HackerNews #ngated

  20. 🔮 Behold, a revolutionary tome unveiling the mystical art of... splitting data sets! 🎩✨ Dive into a world where machine learning geniuses compete in a bizarre contest of who-can-overfit-the-best, and where #benchmarks are the sacred cow 🐄 that everyone loves to hate but won't stop worshipping. Spoiler: it's #groundbreaking, like discovering water is wet. 💧🤯
    mlbenchmarks.org/00-preface.ht #dataScience #machineLearning #overfitting #techHumor #HackerNews #ngated

  21. 🔮 Behold, a revolutionary tome unveiling the mystical art of... splitting data sets! 🎩✨ Dive into a world where machine learning geniuses compete in a bizarre contest of who-can-overfit-the-best, and where #benchmarks are the sacred cow 🐄 that everyone loves to hate but won't stop worshipping. Spoiler: it's #groundbreaking, like discovering water is wet. 💧🤯
    mlbenchmarks.org/00-preface.ht #dataScience #machineLearning #overfitting #techHumor #HackerNews #ngated

  22. 🔮 Behold, a revolutionary tome unveiling the mystical art of... splitting data sets! 🎩✨ Dive into a world where machine learning geniuses compete in a bizarre contest of who-can-overfit-the-best, and where #benchmarks are the sacred cow 🐄 that everyone loves to hate but won't stop worshipping. Spoiler: it's #groundbreaking, like discovering water is wet. 💧🤯
    mlbenchmarks.org/00-preface.ht #dataScience #machineLearning #overfitting #techHumor #HackerNews #ngated

  23. Растягиваем кошек, чтобы избежать переобучения. Аугментация данных в машинном обучении

    Главной проблемой при обучении нейросетей остаётся нехватка качественной информации. Всем моделям глубокого обучения может потребоваться большой объём данных для достижения удовлетворительных результатов. Для успешного обучения модели данные должны быть разнообразными и соответствовать поставленной задаче. В противном случае пользы от такой сети будет мало. Хорошо известно, что нехватка данных легко приводит к переобучению. Но вот беда, трудно предусмотреть и собрать данные, которые покрывали бы все ситуации. Допустим, вы хотите научить систему находить на фото конкретную кошку. Вам потребуются снимки этого животного в самых разных позах — будь то сидя, стоя или обдирающей диван. А если требуется распознавать кошек в принципе, то вариантов становится в разы больше. Видов кошек в природе тысячи, они все разных цветов и размеров. Почему это важно? Представьте, что наш набор данных может содержать изображения кошек и собак. Кошки в наборе смотрят исключительно влево с точки зрения наблюдателя. Неудивительно, что обученная модель может неправильно классифицировать кошек, смотрящих вправо. Поэтому всегда нужно проверять свою выборку на разнообразие. Если данные не подходят под реальные условия, то и задачу решить не получится. Что делать, если у нас дефицит данных?

    habr.com/ru/companies/ruvds/ar

    #ml #машинное+обучение #машинное_обучение #аугментация #аугментация_данных #переобучение #overfitting #синтетические_данные #ruvds_статьи

  24. Растягиваем кошек, чтобы избежать переобучения. Аугментация данных в машинном обучении

    Главной проблемой при обучении нейросетей остаётся нехватка качественной информации. Всем моделям глубокого обучения может потребоваться большой объём данных для достижения удовлетворительных результатов. Для успешного обучения модели данные должны быть разнообразными и соответствовать поставленной задаче. В противном случае пользы от такой сети будет мало. Хорошо известно, что нехватка данных легко приводит к переобучению. Но вот беда, трудно предусмотреть и собрать данные, которые покрывали бы все ситуации. Допустим, вы хотите научить систему находить на фото конкретную кошку. Вам потребуются снимки этого животного в самых разных позах — будь то сидя, стоя или обдирающей диван. А если требуется распознавать кошек в принципе, то вариантов становится в разы больше. Видов кошек в природе тысячи, они все разных цветов и размеров. Почему это важно? Представьте, что наш набор данных может содержать изображения кошек и собак. Кошки в наборе смотрят исключительно влево с точки зрения наблюдателя. Неудивительно, что обученная модель может неправильно классифицировать кошек, смотрящих вправо. Поэтому всегда нужно проверять свою выборку на разнообразие. Если данные не подходят под реальные условия, то и задачу решить не получится. Что делать, если у нас дефицит данных?

    habr.com/ru/companies/ruvds/ar

    #ml #машинное+обучение #машинное_обучение #аугментация #аугментация_данных #переобучение #overfitting #синтетические_данные #ruvds_статьи

  25. Растягиваем кошек, чтобы избежать переобучения. Аугментация данных в машинном обучении

    Главной проблемой при обучении нейросетей остаётся нехватка качественной информации. Всем моделям глубокого обучения может потребоваться большой объём данных для достижения удовлетворительных результатов. Для успешного обучения модели данные должны быть разнообразными и соответствовать поставленной задаче. В противном случае пользы от такой сети будет мало. Хорошо известно, что нехватка данных легко приводит к переобучению. Но вот беда, трудно предусмотреть и собрать данные, которые покрывали бы все ситуации. Допустим, вы хотите научить систему находить на фото конкретную кошку. Вам потребуются снимки этого животного в самых разных позах — будь то сидя, стоя или обдирающей диван. А если требуется распознавать кошек в принципе, то вариантов становится в разы больше. Видов кошек в природе тысячи, они все разных цветов и размеров. Почему это важно? Представьте, что наш набор данных может содержать изображения кошек и собак. Кошки в наборе смотрят исключительно влево с точки зрения наблюдателя. Неудивительно, что обученная модель может неправильно классифицировать кошек, смотрящих вправо. Поэтому всегда нужно проверять свою выборку на разнообразие. Если данные не подходят под реальные условия, то и задачу решить не получится. Что делать, если у нас дефицит данных?

    habr.com/ru/companies/ruvds/ar

    #ml #машинное+обучение #машинное_обучение #аугментация #аугментация_данных #переобучение #overfitting #синтетические_данные #ruvds_статьи

  26. Why Does A.I. Write Like … That?

    Sam Kriss for the New York Times:

    """
    According to the data, post-ChatGPT papers lean more on words like “underscore,” “highlight” and “showcase” than pre-ChatGPT papers [..] And “delve” [..] shot up by 2,700 percent.
    """

    nytimes.com/2025/12/03/magazin

    #EmDash #linguistics #overfitting #ElaraVoss #LLM #NYTimes #SamKriss

  27. Why Does A.I. Write Like … That?

    Sam Kriss for the New York Times:

    """
    According to the data, post-ChatGPT papers lean more on words like “underscore,” “highlight” and “showcase” than pre-ChatGPT papers [..] And “delve” [..] shot up by 2,700 percent.
    """

    nytimes.com/2025/12/03/magazin

  28. Why Does A.I. Write Like … That?

    Sam Kriss for the New York Times:

    """
    According to the data, post-ChatGPT papers lean more on words like “underscore,” “highlight” and “showcase” than pre-ChatGPT papers [..] And “delve” [..] shot up by 2,700 percent.
    """

    nytimes.com/2025/12/03/magazin

    #EmDash #linguistics #overfitting #ElaraVoss #LLM #NYTimes #SamKriss

  29. Why Does A.I. Write Like … That?

    Sam Kriss for the New York Times:

    """
    According to the data, post-ChatGPT papers lean more on words like “underscore,” “highlight” and “showcase” than pre-ChatGPT papers [..] And “delve” [..] shot up by 2,700 percent.
    """

    nytimes.com/2025/12/03/magazin

    #EmDash #linguistics #overfitting #ElaraVoss #LLM #NYTimes #SamKriss

  30. Why Does A.I. Write Like … That?

    Sam Kriss for the New York Times:

    """
    According to the data, post-ChatGPT papers lean more on words like “underscore,” “highlight” and “showcase” than pre-ChatGPT papers [..] And “delve” [..] shot up by 2,700 percent.
    """

    nytimes.com/2025/12/03/magazin

    #EmDash #linguistics #overfitting #ElaraVoss #LLM #NYTimes #SamKriss

  31. Eine große Fehleinschätzung ist, dass #KünstlicheNeuronaleNetzwerke umso besser werden, je komplexer sie sind und je größer der Datensatz ist, mit dem sie trainiert werden. Die aktuell völlig unterschätzte Problematik von #Overfitting & #Overtraining sind potentielle Treiber des nächsten KI-Winters. #justsaying

  32. Eine große Fehleinschätzung ist, dass #KünstlicheNeuronaleNetzwerke umso besser werden, je komplexer sie sind und je größer der Datensatz ist, mit dem sie trainiert werden. Die aktuell völlig unterschätzte Problematik von #Overfitting & #Overtraining sind potentielle Treiber des nächsten KI-Winters. #justsaying

  33. Eine große Fehleinschätzung ist, dass #KünstlicheNeuronaleNetzwerke umso besser werden, je komplexer sie sind und je größer der Datensatz ist, mit dem sie trainiert werden. Die aktuell völlig unterschätzte Problematik von #Overfitting & #Overtraining sind potentielle Treiber des nächsten KI-Winters. #justsaying

  34. Eine große Fehleinschätzung ist, dass #KünstlicheNeuronaleNetzwerke umso besser werden, je komplexer sie sind und je größer der Datensatz ist, mit dem sie trainiert werden. Die aktuell völlig unterschätzte Problematik von #Overfitting & #Overtraining sind potentielle Treiber des nächsten KI-Winters. #justsaying

  35. Eine große Fehleinschätzung ist, dass #KünstlicheNeuronaleNetzwerke umso besser werden, je komplexer sie sind und je größer der Datensatz ist, mit dem sie trainiert werden. Die aktuell völlig unterschätzte Problematik von #Overfitting & #Overtraining sind potentielle Treiber des nächsten KI-Winters. #justsaying

  36. When Dimensionality Hurts: The Role of #LLM Embedding Compression for Noisy Regression Tasks d.repec.org/n?u=RePEc:arx:pape
    "… suggest that the optimal dimensionality is dependent on the signal-to-noise ratio, exposing the necessity of feature compression in high noise environments. The implication of the result is that researchers should consider the #noise of a task when making decisions about the dimensionality of text.

    … findings indicate that sentiment and emotion-based representations do not provide inherent advantages over learned latent features, implying that their previous success in similar tasks may be attributed to #regularisation effects rather than intrinsic informativeness."
    #ML #autoencoders #Overfitting

  37. When Dimensionality Hurts: The Role of #LLM Embedding Compression for Noisy Regression Tasks d.repec.org/n?u=RePEc:arx:pape
    "… suggest that the optimal dimensionality is dependent on the signal-to-noise ratio, exposing the necessity of feature compression in high noise environments. The implication of the result is that researchers should consider the #noise of a task when making decisions about the dimensionality of text.

    … findings indicate that sentiment and emotion-based representations do not provide inherent advantages over learned latent features, implying that their previous success in similar tasks may be attributed to #regularisation effects rather than intrinsic informativeness."
    #ML #autoencoders #Overfitting

  38. When Dimensionality Hurts: The Role of #LLM Embedding Compression for Noisy Regression Tasks d.repec.org/n?u=RePEc:arx:pape
    "… suggest that the optimal dimensionality is dependent on the signal-to-noise ratio, exposing the necessity of feature compression in high noise environments. The implication of the result is that researchers should consider the #noise of a task when making decisions about the dimensionality of text.

    … findings indicate that sentiment and emotion-based representations do not provide inherent advantages over learned latent features, implying that their previous success in similar tasks may be attributed to #regularisation effects rather than intrinsic informativeness."
    #ML #autoencoders #Overfitting

  39. When Dimensionality Hurts: The Role of #LLM Embedding Compression for Noisy Regression Tasks d.repec.org/n?u=RePEc:arx:pape
    "… suggest that the optimal dimensionality is dependent on the signal-to-noise ratio, exposing the necessity of feature compression in high noise environments. The implication of the result is that researchers should consider the #noise of a task when making decisions about the dimensionality of text.

    … findings indicate that sentiment and emotion-based representations do not provide inherent advantages over learned latent features, implying that their previous success in similar tasks may be attributed to #regularisation effects rather than intrinsic informativeness."
    #ML #autoencoders #Overfitting

  40. When Dimensionality Hurts: The Role of #LLM Embedding Compression for Noisy Regression Tasks d.repec.org/n?u=RePEc:arx:pape
    "… suggest that the optimal dimensionality is dependent on the signal-to-noise ratio, exposing the necessity of feature compression in high noise environments. The implication of the result is that researchers should consider the #noise of a task when making decisions about the dimensionality of text.

    … findings indicate that sentiment and emotion-based representations do not provide inherent advantages over learned latent features, implying that their previous success in similar tasks may be attributed to #regularisation effects rather than intrinsic informativeness."
    #ML #autoencoders #Overfitting

  41. 'On the Impact of Hard Adversarial Instances on Overfitting in Adversarial Training', by Chen Liu, Zhichao Huang, Mathieu Salzmann, Tong Zhang, Sabine Süsstrunk.

    jmlr.org/papers/v25/22-0950.ht

    #adversarial #overfitting #robustness

  42. 'On the Impact of Hard Adversarial Instances on Overfitting in Adversarial Training', by Chen Liu, Zhichao Huang, Mathieu Salzmann, Tong Zhang, Sabine Süsstrunk.

    jmlr.org/papers/v25/22-0950.ht

    #adversarial #overfitting #robustness

  43. 'On the Impact of Hard Adversarial Instances on Overfitting in Adversarial Training', by Chen Liu, Zhichao Huang, Mathieu Salzmann, Tong Zhang, Sabine Süsstrunk.

    jmlr.org/papers/v25/22-0950.ht

    #adversarial #overfitting #robustness

  44. 'On the Impact of Hard Adversarial Instances on Overfitting in Adversarial Training', by Chen Liu, Zhichao Huang, Mathieu Salzmann, Tong Zhang, Sabine Süsstrunk.

    jmlr.org/papers/v25/22-0950.ht

    #adversarial #overfitting #robustness

  45. 'On the Impact of Hard Adversarial Instances on Overfitting in Adversarial Training', by Chen Liu, Zhichao Huang, Mathieu Salzmann, Tong Zhang, Sabine Süsstrunk.

    jmlr.org/papers/v25/22-0950.ht

    #adversarial #overfitting #robustness

  46. 1/

    Recent commentary [1]:
    escalating concern over the use of the more powerful #chatbots when they are used to go beyond the #knowledge of the human expert who uses them, rather than for simply pre-processing in a controlled way within the domain of human-expert knowledge.

    1. ⁠What is often called "hallucination/confabulation” (i.e. severe #extrapolation #uncertainty and #overfitting by the chatbot model) is apparently becoming increasingly realistic with a declining human ability to detect it

  47. 1/

    Recent commentary [1]:
    escalating concern over the use of the more powerful #chatbots when they are used to go beyond the #knowledge of the human expert who uses them, rather than for simply pre-processing in a controlled way within the domain of human-expert knowledge.

    1. ⁠What is often called "hallucination/confabulation” (i.e. severe #extrapolation #uncertainty and #overfitting by the chatbot model) is apparently becoming increasingly realistic with a declining human ability to detect it

  48. 1/

    Recent commentary [1]:
    escalating concern over the use of the more powerful #chatbots when they are used to go beyond the #knowledge of the human expert who uses them, rather than for simply pre-processing in a controlled way within the domain of human-expert knowledge.

    1. ⁠What is often called "hallucination/confabulation” (i.e. severe #extrapolation #uncertainty and #overfitting by the chatbot model) is apparently becoming increasingly realistic with a declining human ability to detect it

  49. 1/

    Recent commentary [1]:
    escalating concern over the use of the more powerful #chatbots when they are used to go beyond the #knowledge of the human expert who uses them, rather than for simply pre-processing in a controlled way within the domain of human-expert knowledge.

    1. ⁠What is often called "hallucination/confabulation” (i.e. severe #extrapolation #uncertainty and #overfitting by the chatbot model) is apparently becoming increasingly realistic with a declining human ability to detect it

  50. 1/

    Recent commentary [1]:
    escalating concern over the use of the more powerful #chatbots when they are used to go beyond the #knowledge of the human expert who uses them, rather than for simply pre-processing in a controlled way within the domain of human-expert knowledge.

    1. ⁠What is often called "hallucination/confabulation” (i.e. severe #extrapolation #uncertainty and #overfitting by the chatbot model) is apparently becoming increasingly realistic with a declining human ability to detect it

  51. 'Benign Overfitting of Constant-Stepsize SGD for Linear Regression', by Difan Zou, Jingfeng Wu, Vladimir Braverman, Quanquan Gu, Sham M. Kakade.

    jmlr.org/papers/v24/21-1297.ht

    #overfitting #overparameterized #sgd