AI is undergoing a paradigm shift with the rise of models (eg, BERT, DALL-E, GPT-3) that are
trained on broad data at scale and are adaptable to a wide range of downstream tasks. We …

… We assess a model’s ability to normalize batch effects while preserving biological signal
by evaluating how well it can classify images of treated cells in the out-of-distribution test set. …

Standard training via empirical risk minimization (ERM) can produce models that achieve low
error on average but high error on minority groups, especially in the presence of spurious …

… Classically, we can control the generalization gap with regularization techniques that
constrain the model family’s capacity to fit the training data. In the modern overparameterized …

Almost anywhere you look you can find evidence of stepped-up busi-ness-social sector
interaction. Alongside traditional giving-a local store collecting toys for the needy at Christmas, a …

For machine learning systems to be reliable, we must understand their performance in
unseen, out-of-distribution environments. In this paper, we empirically show that out-of-distribution …

We study why overparameterization—increasing model size well beyond the point of zero
training error—can hurt test error on minority groups despite improving average test error …

… This limits the academic community’s ability to conduct research on autoregressive vision-language …
For ease of comparison to Flamingo, we calculate each OpenFlamingo model’s …

… This phenomenon is ascribable to paramagnetic anisotropy retained by the heme of hemoglobin
S that is polymerized in fiber by deoxygenation. This paper deals with the Orientation of …

… Self-training methods “pseudo-label” unlabeled examples with the model’s own predictions
and then train on them as if they were labeled examples. These methods often also use …