Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revision Previous revision
Next revision
Previous revision
rationalization [2018/08/22 16:29]
admin
rationalization [2019/01/13 20:20] (current)
admin
Line 171: Line 171:
  
 nlike commonly used neural network structures, the structure of the xNN describes the features it learns, via linear projections and univariate functions. These explainability features have the attractive feature of being additive in nature and straightforward to interpret. Whether the network is used as a primary model or a surrogate for a more complex model, the xNN provides straightforward explanations of how the model uses the input features to make predictions. nlike commonly used neural network structures, the structure of the xNN describes the features it learns, via linear projections and univariate functions. These explainability features have the attractive feature of being additive in nature and straightforward to interpret. Whether the network is used as a primary model or a surrogate for a more complex model, the xNN provides straightforward explanations of how the model uses the input features to make predictions.
 +
 +https://​arxiv.org/​abs/​1809.01797 Narrating a Knowledge Base
 +
 +https://​arxiv.org/​abs/​1803.05268 Transparency by Design: Closing the Gap Between Performance and Interpretability in Visual Reasoning
 + ​https://​github.com/​davidmascharka/​tbd-nets https://​towardsdatascience.com/​transparent-reasoning-how-mit-builds-neural-networks-that-can-explain-themselves-3aea291cd9cc
 +
 +https://​arxiv.org/​abs/​1809.06309v1 Commonsense for Generative Multi-Hop Question Answering Tasks
 +
 +https://​arxiv.org/​abs/​1809.07291v1 https://​github.com/​NPoe/​input-optimization-nlp
 +
 +https://​arxiv.org/​pdf/​1805.04833.pdf Hierarchical Neural Story Generation
 +
 +https://​openreview.net/​pdf?​id=rJGgFjA9FQ EXPLAINING ALPHAGO: INTERPRETING CONTEXTUAL
 +EFFECTS IN NEURAL NETWORKS
 +
 +https://​arxiv.org/​pdf/​1804.09160.pdf No Metrics Are Perfect:
 +Adversarial Reward Learning for Visual Storytelling
 +
 +https://​arxiv.org/​abs/​1810.02909v1 On the Art and Science of Machine Learning Explanations
 +
 +https://​arxiv.org/​abs/​1810.03993v1 Model Cards for Model Reporting
 +
 +https://​arxiv.org/​abs/​1810.05680v1 Bottom-up Attention, Models of http://​salicon.net/​
 +
 +https://​github.com/​arviz-devs/​arviz Python package to plot and analyse samples from probabilistic models
 +
 +https://​blog.goodaudience.com/​holy-grail-of-ai-for-enterprise-explainable-ai-xai-6e630902f2a0 ​
 +
 +https://​arxiv.org/​abs/​1809.10736 Controllable Neural Story Generation via Reinforcement Learning
 +
 +We introduce a policy gradient reinforcement learning approach to open story generation that learns to achieve a given narrative goal state. In this work, the goal is for a story to end with a specific type of event, given in advance. ​
 +
 +https://​arxiv.org/​pdf/​1802.07810.pdf Manipulating and Measuring Model Interpretability
 +
 +Participants who were
 +shown a clear model with a small number of features were better able to simulate the model’s predictions. However,
 +contrary to what one might expect when manipulating interpretability,​ we found no significant difference in multiple
 +measures of trust across conditions. Even more surprisingly,​ increased transparency hampered people’s ability to detect
 +when a model has made a sizeable mistake. These findings emphasize the importance of studying how models are
 +presented to people and empirically verifying that interpretable models achieve their intended effects on end users.
 +
 +https://​arxiv.org/​abs/​1703.04730 Understanding Black-box Predictions via Influence Functions
 +
 +
 +https://​christophm.github.io/​interpretable-ml-book/​proto.html
 +
 +