IEEE Transactions on Information Theory 49 14741490

1989

Citations in the corpus

• CPR: Classifier-Projection Regularization for Continual Learning (2020, background): “…observation that our CPR term can be further interpreted in terms of information projection (I-projection) formulations (Cover & Thomas, 2012; Murphy, 2012; Csiszár & Matus, 2003; Walsh & Regalia, 2010; Amari et al., 2001; Csiszár & Matus, 2003; Csiszár & Shields, 2004) found in information theory.”
• Rényi Divergence and Kullback-Leibler Divergence (2012, methodology): “Prior to either of these, the identity (53) below, which forms the heart of the proof, has been used by Csiszár [51].”
• Dynamics-Aware Unsupervised Discovery of Skills (2019, background): “Note, min η ∈ N D KL ( p || η ) , which is the reverse information projection (Csisz´ar & Matus, 2003).”
• Safe Testing (2019, unknown): “Following Barron and Li (1999) (see also (Csiszár and Matus, 2003)), we call P ∗ 0 the Reverse Information Projection (RIPr) of In all examples in this paper, we have P ∗ 0 = P W ∗ 0 : the minimum is achieved and its density integrates to 1 (one can constructspecial H 0 forwhich p ∗ 0…”
• Action and Perception as Divergence Minimization (2020, unknown): “Equation 2 is the reverse KL or information projection used in variational inference (Csiszár and Matus, 2003).”
• Belief Propagation, Dykstra's Algorithm, and Iterated Information Projections (2010, methodology): “Example 4 (Log Partition Function): The conjugate function to the negative Shannon entropy is [27], [28], [31]”
• Precision-Recall Curves Using Information Divergence Frontiers (2019, background): “…over R ∈ M and are thus right information projections onto M (Csiszár and Matus, 2003). are log-convex Then, as exponential families log-convex, from (Csiszár and Matus, 2003, Theorem 1) it follows that for any R ∈ M we have that D KL ( P R ) ≥ D KL ( ˆ P R ) and D KL ( Q R ) ≥ D KL ( ˆ Q R ) ,…”
• Fundamental Limits of Communication With Low Probability of Detection (2015, background): “Then the minimum is well known to be achieved by a distribution from the exponential family connecting Q 0 and Q ∗ [21]: for some λ n ∈ [0 , 1] .”
• Cryptanalysis of the Hillery-Buzek-Berthiaume quantum secret-sharing protocol (2007, background): “This is also the case in quantum cryptography [7, 8, 9, 10, 11, 12].”
• Information Theory and Statistics: A Tutorial (2004, methodology): “1 is a consequence of a general result in [15] for exponential families whose domain of parameters is the whole Rk; the last hypothesis is trivially satisfied in the finite alphabet case.”