La izquierda reniega de la meritocracia porque la ve como una coartada de los privilegios. Pero ¿y si esto fuera un error intelectual y estratégico? ¿Y si estuviera entregando a la derecha una de sus propias ideas más poderosas? En septiembre en librerías: https://t.co/3atVQQa5OF
@DeloraPablo@AvelinoMuleiro Muchas gracias, Pablo. Desconocía que había salido este libro-homenaje. He leído t admirado mucho a Mosterín. Lo compro ahora mismo.
—Papá, ¿puedes darme un ejemplo de palíndromo?
—Yo de todo te doy
—Venga, pues dime uno
—A ti no, bonita
—Pero, ¿en qué quedamos?
—Yo hago yoga hoy
—Mejor le pregunto a mamá
—Ella te dará detalle
#Palíndromos#Maravilloso 😍
Mi director de tesis, Josep Planelles, ha escrito un libro divulgativo en inglés sobre teoría de campos. Si estáis interesados lo podéis descargar gratuitamente en el enlace que comparto. Agradecería difusión para hacerlo llegar a personas interesadas.
➡️ https://t.co/ho9jshZCEU
Backpropagation in the brain? Vectorized teaching signals found in cortical dendrites
Every deep learning algorithm relies on vectorized teaching signals—error information tailored to each neuron—to solve credit assignment: determining which weights should change to improve performance. Backpropagation achieves this but requires a separate backward pass, long considered biologically implausible. How the brain solves credit assignment remains one of the deepest open questions bridging neuroscience and ML.
Theoretical models propose that pyramidal neurons exploit their spatial structure: feedforward inputs arrive near the soma while feedback signals target distant apical dendrites, enabling simultaneous processing of input and teaching signals without a backward phase. Elegant, but lacking direct evidence—until now.
Francioni and coauthors design a neurofeedback brain–computer interface where mice control a visual stimulus by modulating two intermingled populations of layer 5 pyramidal neurons: P+ neurons rotate a grating toward a target, P− neurons rotate it away. This gives experimenters an explicit reward function mapping neural activity to performance, making it possible to define correct teaching signals.
Using two-plane calcium imaging across 14 days of learning, they quantify the somato-dendritic residual—the mismatch between dendritic signal magnitude and what somatic activity predicts. An SVM trained on surrounding network activity can predict whether events are dendritically amplified or attenuated, confirming dendrites carry information absent from somatic firing.
The central result: dendritic signals are neuron-specific and depend on each neuron's causal role. During error reduction, P+ dendrites amplify while P− dendrites attenuate; the pattern reverses during error increase. This is a vectorized teaching signal—each neuron receives error information matched to its contribution, as backpropagation or target propagation would require.
The signals resemble error derivatives rather than absolute error, aligning with difference target propagation over classical backpropagation. Optogenetically silencing apical dendrites abolishes both vectorized signals and learning, establishing dendritic computation as necessary, not merely correlated.
The takeaway: the brain implements a spatial solution to temporal credit assignment. Pyramidal neurons carry feedforward drive at the soma and neuron-specific teaching signals at the dendrites—the first direct biological evidence for dendritic credit assignment models, offering concrete constraints for biologically inspired learning algorithms.
Paper: https://t.co/dg2i4U4NSB
Para celebrar el 2º mes de vida de mi libro...
🪙🪙¡NUEVO SORTEO de “Enséñame a enseñar”!🪙🪙
Para participar:
🔁 Haz RT
❤️ Dale a “Me gusta”
🧵 Sígueme. Lo sortearé el lunes 8 de diciembre.
Dadle caña. ¡Mucha suerte a todos/as!
🎯 𝗠𝗮𝗰𝗵𝗶𝗻𝗲 𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴 + 𝗗𝗲𝗲𝗽 𝗟𝗲𝗮𝗿𝗻𝗶𝗻𝗴 + 𝗔𝗜 — 𝗔𝗹𝗹 𝗖𝗵𝗲𝗮𝘁 𝗦𝗵𝗲𝗲𝘁𝘀!
A complete bundle of cheat sheets from Stanford and MIT, covering everything from ML fundamentals to advanced AI concepts. 🚀
💡 Perfect for students, developers & AI enthusiasts who want quick, clear references while studying or building projects.
FREE Download for the first 500 people!📘
1️⃣ Like & Repost
2️⃣ Comment “AI”
3️⃣ Follow (so I can DM you)
Let’s make learning AI easier, one cheat sheet at a time. 🤖
#MachineLearning #DeepLearning #AI #Giveaway #Stanford #MIT #CheatSheets #AIEducation
Quienes no van a defender la educación y sanidad públicas son ese percentil top20 que se puede pagar la privada y que disfruta de cuidadoras inmigrantes lowcost. Al resto, a los que contaban con vivir limpiando o cuidando ancianos y con un colegio público digno para sus hijos,
1/