Yeah, I didn't quite understand the distinction between autodiff and symbolic differentiation for a while. The analogy of running back a sequence of operations on a tape I found very helpful to understanding, basically never needing to create a closed form representation
September 8, 2026 at 1:10 AM
Yeah, I didn't quite understand the distinction between autodiff and symbolic differentiation for a while. The analogy of running back a sequence of operations on a tape I found very helpful to understanding, basically never needing to create a closed form representation
Reading about autodiff and it's really quite nuts that this was discovered in the freakin 60s (backwards pass in early 70s). What an absurdly clever technique
September 8, 2026 at 12:48 AM
Reading about autodiff and it's really quite nuts that this was discovered in the freakin 60s (backwards pass in early 70s). What an absurdly clever technique
Every gradient in deep learning is matrix calculus. Almost nobody who ships models can do it by hand.
Autodiff made that survivable, not unnecessary. The moment a loss surface misbehaves, or a Jacobian needs interpreting, you are back to the mathematics.
Autodiff made that survivable, not unnecessary. The moment a loss surface misbehaves, or a Jacobian needs interpreting, you are back to the mathematics.
September 7, 2026 at 4:30 PM
Every gradient in deep learning is matrix calculus. Almost nobody who ships models can do it by hand.
Autodiff made that survivable, not unnecessary. The moment a loss surface misbehaves, or a Jacobian needs interpreting, you are back to the mathematics.
Autodiff made that survivable, not unnecessary. The moment a loss surface misbehaves, or a Jacobian needs interpreting, you are back to the mathematics.
Houston Team Maps Quantum Response Between Models
Read more:
https://quantumzeitgeist.com/spectral-autodiff-kernel-expansion-response-mapping/
Read more:
https://quantumzeitgeist.com/spectral-autodiff-kernel-expansion-response-mapping/
Houston Team Maps Quantum Response Between Models
SAKE now delivers spectroscopic responses without recalculating spectra for each new quantum system configuration. Previously reliant on impractical symbolic differentiation as dimensionality increased, accurate predictions are now achieved via automatic differentiation and pathway transport operators. This expansion accurately reproduces projected transport, revealing previously hidden mechanisms within quantum dynamics.
quantumzeitgeist.com
September 2, 2026 at 11:03 PM
Houston Team Maps Quantum Response Between Models
Read more:
https://quantumzeitgeist.com/spectral-autodiff-kernel-expansion-response-mapping/
Read more:
https://quantumzeitgeist.com/spectral-autodiff-kernel-expansion-response-mapping/
SAKE מעביר תגובות ספקטרוסקופיות לא-לינאריות בין מערכות קוונטיות באמצעות הבדלה אוטומטית בשיטה קדימה עם תיאוריית הובלת דוהמל, תוך ביטול חישובים חוזרים יקרים ואיפשור חקירה יעילה של דינמיקה קוונטית מורכבת.
#סימולציהקוונטית #פיזיקהחישובית #חדשות
#סימולציהקוונטית #פיזיקהחישובית #חדשות
SAKE: מסגרת ספקטרלית Autodiff המאיצה סימולציות מערכות קוונטיות
quantumzeitgeist.com
September 2, 2026 at 9:06 PM
SAKE מעביר תגובות ספקטרוסקופיות לא-לינאריות בין מערכות קוונטיות באמצעות הבדלה אוטומטית בשיטה קדימה עם תיאוריית הובלת דוהמל, תוך ביטול חישובים חוזרים יקרים ואיפשור חקירה יעילה של דינמיקה קוונטית מורכבת.
#סימולציהקוונטית #פיזיקהחישובית #חדשות
#סימולציהקוונטית #פיזיקהחישובית #חדשות
SAKE ขนส่งการตอบสนองสเปกโตรสโคปีแบบไม่เชิงเส้นระหว่างระบบควอนตัมโดยใช้การหาอนุพันธ์อัตโนมัติโหมดเดินหน้าพร้อมทฤษฎีการขนส่ง Duhamel ซึ่งขจัดการคำนวณซ้ำที่มีต้นทุนสูงและทำให้สามารถสำรวจพลวัตควอนตัมที่ซับซ้อนได้อย่างมีประสิทธิภาพ
#การจำลองควอนตัม #ฟิสิกส์เชิงคำนวณ #ข่าวสาร
#การจำลองควอนตัม #ฟิสิกส์เชิงคำนวณ #ข่าวสาร
SAKE: กรอบการทำงาน Autodiff สเปกตรัม เร่งการจำลองระบบควอนตัม
iq.fp2.dev
September 2, 2026 at 9:05 PM
SAKE ขนส่งการตอบสนองสเปกโตรสโคปีแบบไม่เชิงเส้นระหว่างระบบควอนตัมโดยใช้การหาอนุพันธ์อัตโนมัติโหมดเดินหน้าพร้อมทฤษฎีการขนส่ง Duhamel ซึ่งขจัดการคำนวณซ้ำที่มีต้นทุนสูงและทำให้สามารถสำรวจพลวัตควอนตัมที่ซับซ้อนได้อย่างมีประสิทธิภาพ
#การจำลองควอนตัม #ฟิสิกส์เชิงคำนวณ #ข่าวสาร
#การจำลองควอนตัม #ฟิสิกส์เชิงคำนวณ #ข่าวสาร
SAKE transportiert nichtlineare spektroskopische Reaktionen zwischen Quantensystemen unter Verwendung von Forward-Mode-Automatischer Differenziation mit Duhamel-Transporttheorie, wodurch teure Neuberechnungen entfallen und eine effiziente Erkundung komplexer Quantendynamik erm...
SAKE: Spektrales Autodiff-Framework Beschleunigt Quantensystem-Simulationen
iq.fp2.dev
September 2, 2026 at 9:05 PM
SAKE transportiert nichtlineare spektroskopische Reaktionen zwischen Quantensystemen unter Verwendung von Forward-Mode-Automatischer Differenziation mit Duhamel-Transporttheorie, wodurch teure Neuberechnungen entfallen und eine effiziente Erkundung komplexer Quantendynamik erm...
SAKE transporte les réponses spectroscopiques non linéaires entre les systèmes quantiques en utilisant la différentiation automatique en mode avant avec la théorie du transport de Duhamel, éliminant les recalculs coûteux et permettant une exploration efficace de la dynamique q...
SAKE : Cadre Autodiff Spectral Accélère les Simulations de Systèmes Quantiques
quantumzeitgeist.com
September 2, 2026 at 9:04 PM
SAKE transporte les réponses spectroscopiques non linéaires entre les systèmes quantiques en utilisant la différentiation automatique en mode avant avec la théorie du transport de Duhamel, éliminant les recalculs coûteux et permettant une exploration efficace de la dynamique q...
SAKE transports nonlinear spectroscopic responses between quantum systems using forward-mode automatic differentiation with Duhamel transport theory, eliminating expensive recalculations and enabling efficient exploration of complex quantum dynamics.
#QuantumSimulation #ComputationalPhysics #News
#QuantumSimulation #ComputationalPhysics #News
SAKE: Spectral Autodiff Framework Accelerates Quantum System Simulations
quantumzeitgeist.com
September 2, 2026 at 9:03 PM
SAKE transports nonlinear spectroscopic responses between quantum systems using forward-mode automatic differentiation with Duhamel transport theory, eliminating expensive recalculations and enabling efficient exploration of complex quantum dynamics.
#QuantumSimulation #ComputationalPhysics #News
#QuantumSimulation #ComputationalPhysics #News
Every gradient in deep learning is matrix calculus. Almost nobody who ships models can do it by hand.
Autodiff made that survivable, not unnecessary. The moment a loss surface misbehaves, or a Jacobian needs interpreting, you are back to the mathematics.
Autodiff made that survivable, not unnecessary. The moment a loss surface misbehaves, or a Jacobian needs interpreting, you are back to the mathematics.
August 28, 2026 at 4:30 PM
Every gradient in deep learning is matrix calculus. Almost nobody who ships models can do it by hand.
Autodiff made that survivable, not unnecessary. The moment a loss surface misbehaves, or a Jacobian needs interpreting, you are back to the mathematics.
Autodiff made that survivable, not unnecessary. The moment a loss surface misbehaves, or a Jacobian needs interpreting, you are back to the mathematics.
Desde gradientes hasta ChatGPT: armá tu propio LLM
¿Querés entender un LLM de verdad? Este curso gratuito va desde gradientes hasta ChatGPT en 20 módulos usando solo tu MacBook, sin GPUs en la nube
#llm #chatgpt #cursogratuito #transformer #deeplearning
¿Querés entender un LLM de verdad? Este curso gratuito va desde gradientes hasta ChatGPT en 20 módulos usando solo tu MacBook, sin GPUs en la nube
#llm #chatgpt #cursogratuito #transformer #deeplearning
Desde gradientes hasta ChatGPT: armá tu propio LLM
Curso de autoestudio con 20 módulos: escribís autodiff, transformer, SFT y RAG, y terminás con un asistente de chat hecho por vos, todo en una MacBook M-series y sin costo de cómputo.
blog.donweb.com
August 25, 2026 at 11:23 PM
Desde gradientes hasta ChatGPT: armá tu propio LLM
¿Querés entender un LLM de verdad? Este curso gratuito va desde gradientes hasta ChatGPT en 20 módulos usando solo tu MacBook, sin GPUs en la nube
#llm #chatgpt #cursogratuito #transformer #deeplearning
¿Querés entender un LLM de verdad? Este curso gratuito va desde gradientes hasta ChatGPT en 20 módulos usando solo tu MacBook, sin GPUs en la nube
#llm #chatgpt #cursogratuito #transformer #deeplearning
Same's true of NNs - you no longer have to calculate autodiff layers by hand, you just import jax/torch and stack layers. Or Bayes - no need to write your own MCMC, you just use Stan/PyMC/(etc)
August 25, 2026 at 6:07 PM
Same's true of NNs - you no longer have to calculate autodiff layers by hand, you just import jax/torch and stack layers. Or Bayes - no need to write your own MCMC, you just use Stan/PyMC/(etc)
Watched Bob Carpenter's StanCon talk and decided the Stan stack deserved some love. PRs incoming to stan-math, stanc3, bridgestan + walnutpie: −15% walltime per gradient, a silent autodiff blowup on degenerate eigenvalues (ESS 29→411 when fixed), and a small pile of bugs 👀 :3
#rstats #stan
#rstats #stan
August 23, 2026 at 11:33 AM
מסגרת SAKE משלבת הבחנה אוטומטית עם תורת התחבורה של Duhamel להעברה יעילה של ספקטרוסקופיה קוונטית לא-ליניארית בין מודלים שכנים, מה שמאפשר הערכת פרמטרים מהירה וחושפת מנגנוני ערבוב נתיבים המוסתרים בספקטרה.
#ספקטרוסקופיהקוונטית #הבחנהאוטומטית #מחקר
#ספקטרוסקופיהקוונטית #הבחנהאוטומטית #מחקר
התרחבות ליבה Autodiff ספקטרלית לתחבורת תגובה קוונטית יעילה
arxiv.org
August 21, 2026 at 4:25 AM
מסגרת SAKE משלבת הבחנה אוטומטית עם תורת התחבורה של Duhamel להעברה יעילה של ספקטרוסקופיה קוונטית לא-ליניארית בין מודלים שכנים, מה שמאפשר הערכת פרמטרים מהירה וחושפת מנגנוני ערבוב נתיבים המוסתרים בספקטרה.
#ספקטרוסקופיהקוונטית #הבחנהאוטומטית #מחקר
#ספקטרוסקופיהקוונטית #הבחנהאוטומטית #מחקר
Das SAKE-Framework kombiniert automatische Differenziation mit der Duhamel-Transporttheorie, um nichtlineare Quantenspektroskopie effizient zwischen benachbarten Modellen zu transportieren und ermöglicht so schnelle Parameterschätzung sowie die Offenbarung von Pfadmischungsmec...
Spektrale Autodiff-Kern-Expansion für effizienten Quantenresponse-Transport
arxiv.org
August 21, 2026 at 4:23 AM
Das SAKE-Framework kombiniert automatische Differenziation mit der Duhamel-Transporttheorie, um nichtlineare Quantenspektroskopie effizient zwischen benachbarten Modellen zu transportieren und ermöglicht so schnelle Parameterschätzung sowie die Offenbarung von Pfadmischungsmec...
Le cadre SAKE combine la différentiation automatique à la théorie du transport de Duhamel pour transporter efficacement la spectroscopie quantique non-linéaire entre les modèles voisins, permettant l'estimation rapide des paramètres et révélant les mécanismes de mélange de voi...
Expansion Spectrale du Noyau Autodiff pour le Transport Efficace de Réponse Quantique
arxiv.org
August 21, 2026 at 4:22 AM
Le cadre SAKE combine la différentiation automatique à la théorie du transport de Duhamel pour transporter efficacement la spectroscopie quantique non-linéaire entre les modèles voisins, permettant l'estimation rapide des paramètres et révélant les mécanismes de mélange de voi...
SAKE framework combines automatic differentiation with Duhamel transport theory to efficiently transport nonlinear quantum spectroscopy between neighboring models, enabling fast parameter estimation and revealing pathway mixing mechanisms hidden in spectra.
Spectral Autodiff Kernel Expansion for Efficient Quantum Response Transport
arxiv.org
August 21, 2026 at 4:21 AM
SAKE framework combines automatic differentiation with Duhamel transport theory to efficiently transport nonlinear quantum spectroscopy between neighboring models, enabling fast parameter estimation and revealing pathway mixing mechanisms hidden in spectra.
Every gradient in deep learning is matrix calculus. Almost nobody who ships models can do it by hand.
Autodiff made that survivable, not unnecessary. The moment a loss surface misbehaves, or a Jacobian needs interpreting, you are back to the mathematics.
Autodiff made that survivable, not unnecessary. The moment a loss surface misbehaves, or a Jacobian needs interpreting, you are back to the mathematics.
August 19, 2026 at 1:30 PM
Every gradient in deep learning is matrix calculus. Almost nobody who ships models can do it by hand.
Autodiff made that survivable, not unnecessary. The moment a loss surface misbehaves, or a Jacobian needs interpreting, you are back to the mathematics.
Autodiff made that survivable, not unnecessary. The moment a loss surface misbehaves, or a Jacobian needs interpreting, you are back to the mathematics.
The real win: the gradient becomes one sparse solve of the fixed-point condition instead of backprop through the whole iterative contraction. The environment's near-degenerate fixed point is exactly what makes autodiff unstable in PEPS — this is what makes 2D variational TNs actually usable.
August 17, 2026 at 11:54 AM
The real win: the gradient becomes one sparse solve of the fixed-point condition instead of backprop through the whole iterative contraction. The environment's near-degenerate fixed point is exactly what makes autodiff unstable in PEPS — this is what makes 2D variational TNs actually usable.
Exploration around how to organize a space of generative models so it remains differentiable: chatgpt.com/share/6a81dd...
It also implemented an hamiltonian monte carlo "hello world" for me using Rust std::autodiff (while teaching me Rust).
It also implemented an hamiltonian monte carlo "hello world" for me using Rust std::autodiff (while teaching me Rust).
August 17, 2026 at 12:56 AM
Exploration around how to organize a space of generative models so it remains differentiable: chatgpt.com/share/6a81dd...
It also implemented an hamiltonian monte carlo "hello world" for me using Rust std::autodiff (while teaching me Rust).
It also implemented an hamiltonian monte carlo "hello world" for me using Rust std::autodiff (while teaching me Rust).
Every gradient in deep learning is matrix calculus. Almost nobody who ships models can do it by hand.
Autodiff made that survivable, not unnecessary. The moment a loss surface misbehaves, or a Jacobian needs interpreting, you are back to the mathematics.
Autodiff made that survivable, not unnecessary. The moment a loss surface misbehaves, or a Jacobian needs interpreting, you are back to the mathematics.
August 15, 2026 at 4:30 PM
Every gradient in deep learning is matrix calculus. Almost nobody who ships models can do it by hand.
Autodiff made that survivable, not unnecessary. The moment a loss surface misbehaves, or a Jacobian needs interpreting, you are back to the mathematics.
Autodiff made that survivable, not unnecessary. The moment a loss surface misbehaves, or a Jacobian needs interpreting, you are back to the mathematics.
This is so niche.
I found this on my phone from 2023.
I don't remember it, 2019-2024 I was indeed talking a lot to both when I sZ working on Autodiff
I found this on my phone from 2023.
I don't remember it, 2019-2024 I was indeed talking a lot to both when I sZ working on Autodiff
August 15, 2026 at 8:27 AM
This is so niche.
I found this on my phone from 2023.
I don't remember it, 2019-2024 I was indeed talking a lot to both when I sZ working on Autodiff
I found this on my phone from 2023.
I don't remember it, 2019-2024 I was indeed talking a lot to both when I sZ working on Autodiff
Pierre Granger: MANGO: An Autodiff Neutrino Oscillation Engine for Differentiable Analysis Pipelines https://arxiv.org/abs/2608.13429 https://arxiv.org/pdf/2608.13429 https://arxiv.org/html/2608.13429
August 14, 2026 at 6:53 AM
Pierre Granger: MANGO: An Autodiff Neutrino Oscillation Engine for Differentiable Analysis Pipelines https://arxiv.org/abs/2608.13429 https://arxiv.org/pdf/2608.13429 https://arxiv.org/html/2608.13429
#AutoDifferentiation in #MachineLearning Algebra! #BigData #Analytics #DataScience #AI #MachineLearning #IoT #IIoT #Python #RStats #TensorFlow #JavaScript #ReactJS #Mathematics #Serverless #DataScientist #Linux #Programming #Coding #100DaysofCode
geni.us/AutoDiff-ML-...
geni.us/AutoDiff-ML-...
August 10, 2026 at 6:26 AM
is there a non-python implementation of autodiff?
August 8, 2026 at 4:39 AM
is there a non-python implementation of autodiff?