# First is Better Than Last for Language Data Influence

However, we observe that since the activation connected to the last layer of weights contains "shared logic", the data influenced calculated via the last layer weights prone to a cancellation effect'', where the data influence of different examples have large magnitude that contradicts each other.

1

# Reciprocity in Machine Learning

no code implementations19 Feb 2022,

Are these contributions (outflows of influence) and benefits (inflows of influence) reciprocal?

# The Penalty Imposed by Ablated Data Augmentation

There is a set of data augmentation techniques that ablate parts of the input at random.

# Attribution in Scale and Space

Third, it eliminates the need for a 'baseline' parameter for Integrated Gradients  for perception tasks.

858

# Estimating Training Data Influence by Tracing Gradient Descent

We introduce a method called TracIn that computes the influence of a training example on a prediction made by the model.

3,714

# The many Shapley values for model explanation

In this paper, we use the axiomatic approach to study the differences between some of the many operationalizations of the Shapley value for attribution, and propose a technique called Baseline Shapley (BShap) that is backed by a proper uniqueness result.

# How Important is a Neuron

Informally, the conductance of a hidden unit of a deep network is the flow of attribution via this hidden unit.

# A Note about: Local Explanation Methods for Deep Neural Networks lack Sensitivity to Parameter Values

no code implementations11 Jun 2018,

Local explanation methods, also known as attribution methods, attribute a deep network's prediction to its input (cf.

# How Important Is a Neuron?

Informally, the conductance of a hidden unit of a deep network is the \emph{flow} of attribution via this hidden unit.

3,714

# Did the Model Understand the Question?

Our strongest attacks drop the accuracy of a visual question answering model from $61. 1\%$ to $19\%$, and that of a tabular question answering model from $33. 5\%$ to $3. 3\%$.

514

# It was the training data pruning too!

The large impact on the performance of the KDG model suggests that the pruning may be a useful pre-processing step in training other semantic parsers as well.

# Abductive Matching in Question Answering

We study question-answering over semi-structured data.

# Axiomatic Attribution for Deep Networks

We study the problem of attributing the prediction of a deep network to its input features, a problem previously studied by several other works.

3,714