no code implementations • 1 May 2024 • Adi Hendel, Meir Feder
In this paper, we consider PAC learning using a somewhat different tradeoff, the error exponent - a well established analysis method in Information Theory - which describes the exponential behavior of the probability that the risk will exceed a certain threshold as function of the sample size.
no code implementations • 14 Jun 2023 • Shahar Stein Ioushua, Inbar Hasidim, Ofer Shayevitz, Meir Feder
Learning algorithms that divide the data into batches are prevalent in many machine-learning applications, typically offering useful trade-offs between computational efficiency and performance.
no code implementations • 17 Jun 2022 • Koby Bibas, Meir Feder
In the context of online prediction where the min-max solution is the Normalized Maximum Likelihood (NML), it has been suggested to use NML with ``luckiness'': A prior-like function is applied to the hypothesis class, which reduces its effective size.
1 code implementation • NeurIPS 2021 • Koby Bibas, Meir Feder, Tal Hassner
Furthermore, we describe how to efficiently apply the derived pNML regret to any pretrained deep NN, by employing the explicit pNML for the last layer, followed by the softmax function.
Out-of-Distribution Detection Out of Distribution (OOD) Detection
no code implementations • 4 Sep 2021 • Uriya Pesso, Koby Bibas, Meir Feder
Specifically, our defense performs adversarial targeted attacks according to different hypotheses, where each hypothesis assumes a specific label for the test sample.
no code implementations • 14 Feb 2021 • Koby Bibas, Meir Feder
Modern machine learning models do not obey this paradigm: They produce an accurate prediction even with a perfect fit to the training set.
no code implementations • 29 Jan 2021 • Meir Feder, Yury Polyanskiy
The well-specified case corresponds to an additional assumption that the data-generating distribution belongs to the hypothesis class as well.
no code implementations • 20 Nov 2020 • Yaniv Fogel, Tal Shapira, Meir Feder
This approach has yields a learnability measure that can also be interpreted as a stability measure.
no code implementations • NeurIPS 2020 • Assaf Dauber, Meir Feder, Tomer Koren, Roi Livni
The notion of implicit bias, or implicit regularization, has been suggested as a means to explain the surprising generalization ability of modern-days overparameterized learning algorithms.
no code implementations • 25 Sep 2019 • Uriya Pesso, Koby Bibas, Meir Feder
In particular, we follow the recently suggested Predictive Normalized Maximum Likelihood (pNML) scheme for universal learning, whose goal is to optimally compete with a reference learner that knows the true label of the test sample but is restricted to use a learner from a given hypothesis class.
3 code implementations • 12 May 2019 • Koby Bibas, Yaniv Fogel, Meir Feder
Linear regression is a classical paradigm in statistics.
1 code implementation • 28 Apr 2019 • Koby Bibas, Yaniv Fogel, Meir Feder
Finally, we extend the pNML to a ``twice universal'' solution, that provides universality for model class selection and generates a learner competing with the best one from all model classes.
no code implementations • 22 Dec 2018 • Yaniv Fogel, Meir Feder
Universal supervised learning is considered from an information theoretic point of view following the universal prediction approach, see Merhav and Feder (1998).
no code implementations • 31 Oct 2018 • Amichai Painsky, Meir Feder, Naftali Tishby
In this work we introduce an information-theoretic compressed representation framework for the non-linear CCA problem (CRCCA), which extends the classical ACE approach.
no code implementations • 16 Sep 2018 • Amichai Painsky, Saharon Rosset, Meir Feder
Importantly, we show that the overhead of our suggested algorithm (compared with the lower bound) typically decreases, as the scale of the problem grows.
no code implementations • 6 Jul 2018 • Amichai Painsky, Meir Feder
Estimating a large alphabet probability distribution from a limited number of samples is a fundamental problem in machine learning and statistics.
no code implementations • 23 Oct 2013 • Ronen Dar, Meir Feder, Antonio Mecozzi, Mark Shtaif
Through a series of extensive system simulations we show that all of the previously not understood discrepancies between the Gaussian noise (GN) model and simulations can be attributed to the omission of an important, recently reported, fourth-order noise (FON) term, that accounts for the statistical dependencies within the spectrum of the interfering channel.
Optics