Deep learning algorithms are well-known to have a propensity for fitting the training data very well and often fit even outliers and mislabeled data points. Such fitting requires memorization of training data labels, a phenomenon that has attracted significant research interest but has not been given a compelling explanation so far. A recent work of Feldman (2019) proposes a theoretical explanation for this phenomenon based on a combination of two insights. First, natural image and data distributions are (informally) known to be long-tailed, that is have a significant fraction of rare and atypical examples. Second, in a simple theoretical model such memorization is necessary for achieving close-to-optimal generalization error when the data distribution is long-tailed.

Weaknesses: The idea of applying Shapley values for the understanding of deep neural networks is not new. Several works, such as Lundberg et al., 2017, have already discussed the theoretical motivation for using Shapley values as an attribution method to rank the importance of the input features. Lundberg et al., 2017 also proposed approximations like KernelSHAP and DeepSHAP, which are not compared to TMAB-Shapley. Besides this line of works, the idea of using Shapley values to rank the internal neurons has been proposed by the Stier et al., 2018 (cited) and Florin Leon, 2014 (not cited) in the context of pruning. Finally, Ancona et al., 2019 (not cited) proposed an approximation technique for Shapley values tailored for deep neural networks.

Weaknesses: I would like to see some clarification on the long tail theory. If the value of mem(A,S,i_1,...,i_k) is high, perhaps we can still call this phenomenon "memorization." If so, then memorization phenomenon is not just limited to long tails. Then, it seems to me the claim in [12] that memorization is needed due to long tail may not be showing a bigger picture. The paper mentions that very high influence scores are due to near duplicates in the training and test examples.

The reviews feel that the issues are interesting and the contributions are sufficient for acceptance. However, there are serious suggestions for improvements in the experiments. It seems the paper is suggestive, but not definitive, on the long tail hypothesis.

Deep learning algorithms are well-known to have a propensity for fitting the training data very well and often fit even outliers and mislabeled data points. Such fitting requires memorization of training data labels, a phenomenon that has attracted significant research interest but has not been given a compelling explanation so far. A recent work of Feldman (2019) proposes a theoretical explanation for this phenomenon based on a combination of two insights. First, natural image and data distributions are (informally) known to be long-tailed, that is have a significant fraction of rare and atypical examples. Second, in a simple theoretical model such memorization is necessary for achieving close-to-optimal generalization error when the data distribution is long-tailed.

Consider a robot wandering around an unfamiliar environment. The robot's task is to con(cid:173) struct an internal model of its environment. The heart of this algorithm is a clever representation of the environment called an update graph. We have developed a connectionist implementation of the update graph using a highly-specialized network architecture. The network has the additional strength that it can accommodate stochastic environments.

To do this, Lipson and colleagues have designed a machine learning algorithm capable of studying physical phenomena by "watching" videos, such as the swing of a double pendulum or the flicker of a flame, and producing the number of variables needed to explain the action. For known systems, the algorithm was able to predict the correct number of variables within 1 value (e.g. The findings were published last week in a study titled "Automated discovery of fundamental variables hidden in experimental data" in the journal Nature Computational Science.

In this blog post, we introduce Domino, a new approach for discovering systematic errors made by machine learning models. We also discuss a framework for quantitatively evaluating methods like Domino. Machine learning models that achieve high overall accuracy often make systematic errors on coherent slices of validation data. A slice is a set of data samples that share a common characteristic. As an example, in large image datasets, photos of vintage cars comprise a slice (i.e.

DALLE-2 is a deep generative model that takes a text caption and generates images that match the given text. However, it has its limitations. Sample image generated using DALL·E 2. Image credit: OpenAI Researchers discover that this text is not random but reveals a hidden vocabulary that the model seems to have developed internally. Researchers find that words that sound gibberish for humans may have a meaning for DALLE-2; for example, Vicootes means vegetables. Researchers notice that a system behaving in unpredictable ways may cause security concerns.

While the insurance industry has struggled with adopting new digital paths, insurance companies have noticed the growth of highly digitized new market entrants, some of which they may want to emulate. However, as we have seen from past experiences, the copycat approach is often the least successful as only tested breakthrough ideas can yield high financial returns. In just 6 years and despite the pandemic, InsurTech startups are now counting at least 23 well-known unicorn startups within their ranks or insurance companies with a billion-dollar valuation. These digital insurers and service providers use digital technologies to design, deliver, and price insurance offers uniquely. We estimate that the total valuation of our selected 23 companies today represents over $60 billion in valuation within an 850 strong global portfolio worth $2.4 Trillion.