Axiomatic Attribution for Deep Networks
- Mukund Sundararajan, Ankur Taly, Qiqi Yan
- Computer ScienceInternational Conference on Machine Learning
- 4 March 2017
We study the problem of attributing the prediction of a deep network to its input features, a problem previously studied by several other works. We identify two fundamental axioms— Sensitivity and…
Revenue maximization with a single sample
- Peerapong Dhangwatnotai, T. Roughgarden, Qiqi Yan
- Economics, MathematicsACM Conference on Economics and Computation
- 7 June 2010
This work designs and analyzes approximately revenue-maximizing auctions in general single-parameter settings and gives an auction that, for every environment and unknown valuation distributions, has expected revenue at least a constant fraction of the expected optimal welfare.
Online bipartite matching with random arrivals: an approach based on strongly factor-revealing LPs
- Mohammad Mahdian, Qiqi Yan
- Computer ScienceSymposium on the Theory of Computing
- 6 June 2011
This paper studies the ranking algorithm in the random arrivals model, and shows that it has a competitive ratio of at least 0.696, beating the 1-1/e ≈ 0.632 barrier in the adversarial model.
Mechanism design via correlation gap
- Qiqi Yan
- Computer ScienceACM-SIAM Symposium on Discrete Algorithms
- 11 August 2010
A theoretical explanation of how certain SPMs can approximate the optimal mechanism by a good constant factor is given, and tight analysis of a greedy-based SPM of Chawla et al. for several environments is given.
Gradients of Counterfactuals
- Mukund Sundararajan, Ankur Taly, Qiqi Yan
- Computer ScienceArXiv
- 8 November 2016
This work proposes to examine interior gradients, which are gradients of counterfactual inputs constructed by scaling down the original input, and applies this method to the GoogleNet architecture, as well as a ligand-based virtual screening network with categorical features and an LSTM based language model for the Penn Treebank dataset.
How Important Is a Neuron?
- Kedar Dhamdhere, Mukund Sundararajan, Qiqi Yan
- Computer ScienceInternational Conference on Learning…
- 30 May 2018
The notion of conductance is introduced to extend the notion of attribution to the understanding of the importance of hidden units in a deep network, or over a set of inputs.
Lower Bounds for Complementation of omega-Automata Via the Full Automata Technique
- Qiqi Yan
- Computer ScienceLog. Methods Comput. Sci.
- 10 July 2006
A new lower bound technique for the state complexity of transformations of automata is introduced and an Ω((0.76n)n) lower bound for Buchi complementation is proved, which also holds for almost every complementation and determinization transformation of nondeterministic ω-automata.
Lower Bounds for Complementation of ω-Automata Via the Full Automata Technique
- Qiqi Yan
- Computer Science
- 2006
A new lower bound technique is introduced for the state complexity of transformations of automata and an Ω((0.76n) n ) lower bound for Buchi complementation is proved, which also holds for almost every complementation and determinization transformation of nondeterministicπ-automata.
Robust mechanisms for risk-averse sellers
- Mukund Sundararajan, Qiqi Yan
- EconomicsACM Conference on Economics and Computation
- 11 February 2010
This paper identifies good mechanisms for risk-averse sellers and identifies a posted-price mechanism called the Hedge mechanism, which gives a universal constant factor approximation and establishes interesting properties of regular distributions and monotone hazard rate distributions.
Analyza: Exploring Data with Conversation
- Kedar Dhamdhere, K. McCurley, Ralfi Nahmias, Mukund Sundararajan, Qiqi Yan
- Computer ScienceInternational Conference on Intelligent User…
- 7 March 2017
Analyza, a system that helps lay users explore data and discuss the key design decisions in implementing this system, including how to mix structured and natural language modalities, how to use conversation to disambiguate and simplify querying, and how to efficiently curate the data.
...
...