Universal Decoding for the Typical Random Code and for the Expurgated Code

Tamir, Ran; Merhav, Neri

doi:10.1109/isit50566.2022.9834903

Cited by 2 publications

(6 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Nevertheless, identification proceeded by sequential detection would take on average 3.000 queries (24 times more w.r.t. detection) as foreseen by (10).…”

Section: Identification (I F a B K)mentioning

confidence: 96%

“…(10). For instance, for singleton family, E(|F|) = 1 and the rate equals 0.002 under top-1, whereas the rate in (10) cannot be lower than 0.5 since we need at least one query to discard a hypothesis, i.e. E(L neg j ) ≥ 1.…”

Section: Identificationmentioning

confidence: 99%

“…Comparing Figures 2 and 3, two times more queries are necessary for identifying a family rather than detecting it. It is possible to identify a model quickly with at most five benign queries which are a lot less than the sequential procedure (10). Identification is a harder task than detection to a small extent.…”

Section: Identificationmentioning

confidence: 99%

“…Resorting to the empirical mutual information to decode transmitted messages in digital communication is known as Maximum Mutual Information (MMI), recently proven universally optimal [10].…”

Section: A Discriminative Distancementioning

confidence: 99%

“…8. C Proof of Equation (10) Suppose that Bob selects b ∈ F j . Alice makes a random permutation σ of the n F families and sequentially tests them.…”

Section: A Notationsmentioning

confidence: 99%

See 4 more Smart Citations

FBI: Fingerprinting models with Benign Inputs

Maho,

Furon,

Merrer

2022

Preprint

View full text Add to dashboard Cite

Recent advances in the fingerprinting of deep neural networks detect instances of models, placed in a black-box interaction scheme. Inputs used by the fingerprinting protocols are specifically crafted for each precise model to be checked for. While efficient in such a scenario, this nevertheless results in a lack of guarantee after a mere modification (e.g. retraining, quantization) of a model.This paper tackles the challenges to propose i) fingerprinting schemes that are resilient to significant modifications of the models, by generalizing to the notion of model families and their variants, ii) an extension of the fingerprinting task encompassing scenarios where one wants to fingerprint not only a precise model (previously referred to as a detection task) but also to identify which model family is in the black-box (identification task).We achieve both goals by demonstrating that benign inputs, that are unmodified images, for instance, are sufficient material for both tasks. We leverage an information-theoretic scheme for the identification task. We devise a greedy discrimination algorithm for the detection task. Both approaches are experimentally validated over an unprecedented set of more than 1,000 networks.

show abstract

“…Nevertheless, identification proceeded by sequential detection would take on average 3.000 queries (24 times more w.r.t. detection) as foreseen by (10).…”

Section: Identification (I F a B K)mentioning

confidence: 96%

Section: Identificationmentioning

confidence: 99%

Section: Identificationmentioning

confidence: 99%

Section: A Discriminative Distancementioning

confidence: 99%

“…8. C Proof of Equation (10) Suppose that Bob selects b ∈ F j . Alice makes a random permutation σ of the n F families and sequentially tests them.…”

Section: A Notationsmentioning

confidence: 99%

See 3 more Smart Citations

FBI: Fingerprinting models with Benign Inputs

Maho,

Furon,

Merrer

2022

Preprint

View full text Add to dashboard Cite

show abstract

Trade-offs between Error Exponents and Excess-Rate Exponents of Typical Slepian–Wolf Codes

Tamir

Merhav

2021

Entropy

View full text Add to dashboard Cite

Typical random codes (TRCs) in a communication scenario of source coding with side information in the decoder is the main subject of this work. We study the semi-deterministic code ensemble, which is a certain variant of the ordinary random binning code ensemble. In this code ensemble, the relatively small type classes of the source are deterministically partitioned into the available bins in a one-to-one manner. As a consequence, the error probability decreases dramatically. The random binning error exponent and the error exponent of the TRCs are derived and proved to be equal to one another in a few important special cases. We show that the performance under optimal decoding can be attained also by certain universal decoders, e.g., the stochastic likelihood decoder with an empirical entropy metric. Moreover, we discuss the trade-offs between the error exponent and the excess-rate exponent for the typical random semi-deterministic code and characterize its optimal rate function. We show that for any pair of correlated information sources, both error and excess-rate probabilities exponential vanish when the blocklength tends to infinity.

show abstract

Universal Decoding for the Typical Random Code and for the Expurgated Code

Cited by 2 publications

References 25 publications

FBI: Fingerprinting models with Benign Inputs

FBI: Fingerprinting models with Benign Inputs

Trade-offs between Error Exponents and Excess-Rate Exponents of Typical Slepian–Wolf Codes

Contact Info

Product

Resources

About