Descripción del proyecto
Descubrir cómo funciona realmente el aprendizaje profundo
El aprendizaje profundo (AP o DL, por sus siglas en inglés) simula una red neuronal utilizando múltiples capas para aprender de forma autónoma a partir de los datos. A pesar de ser eficaz en muchas aplicaciones, se desconoce su fundamento teórico. El proyecto UnderstandingDL, financiado por el Consejo Europeo de Investigación, adoptará un triple enfoque. En primer lugar, desarrollará nuevos modelos aprendibles mediante el descenso del gradiente estocástico, un algoritmo de optimización utilizado para entrenar redes neuronales. Ello puede producir nuevos algoritmos de AP que se comprenden a nivel teórico. En segundo lugar, se investigará la complejidad de la muestra de diferentes tipos de redes neuronales y la intensidad de las conexiones de los nodos. En consecuencia, eso debería explicar cómo es posible que el AP pueda tener menos ejemplos que los parámetros y, sin embargo, realizar la generalización de forma satisfactoria. Por último, se investigará la funcionalidad de las redes neuronales, incluidas las ventajas relativas a la profundidad y los efectos de los datos corruptos de forma deliberada (los denominados «ejemplos adversos»).
Objetivo
While extremely successful, deep learning (DL) still lacks a solid theoretical foundation.
In the last 5 years the PI focused almost entirely on DL theory, yielding a strong publication record with 7 papers at NeurIPS (the leading ML conference), including 2 spotlights (top 3% of submitted papers) and one oral (top 1%), 2 papers at ICLR (the leading DL conference), and 1 paper at COLT (the leading ML theory conference). These results are amongst the first that break a 20 years hiatus in NN theory, thereby giving some hope for a solid deep learning theory. This includes 1) the first poly-time learnability result for non-trivial function class by SGD on NN, 2) the first such result with near optimal rate, 3) new methodology to bound the sample complexity of NN, that established the first sample complexity bound that is sublinear in the number of parameters, under norm constraints that are valid in practice, 4) an explanation to the phenomena of adversarial examples.
We plan to go far beyond these and other results, and to build a coherent theory for DL, addressing all three pillars of learning theory:
Optimization: We plan to investigate the success of SGD in finding a good model, arguably the greatest mystery of modern deep learning. Specifically our goal is to understand what models are learnable by SGD on neural networks. To this end, we plan to come up with a new class of models that can potentially lead to new deep learning algorithms, with a solid theory behind them.
Statistical Complexity: We plan to crack the second great mystery of modern deep learning, which is their ability to generalize with fewer examples than parameters. Our plan is to investigate the sample complexity of classes of neural networks that are defined by bounds on the weights’ magnitude.
Representation: We plan to investigate functions that can be realized by NN. This includes classical questions such as the benefits of depth, as well as more modern aspects such as adversarial examples.
Programa(s)
- HORIZON.1.1 - European Research Council (ERC) Main Programme
Régimen de financiación
HORIZON-ERC - HORIZON ERC GrantsInstitución de acogida
91904 Jerusalem
Israel