Refine
Year of publication
Document Type
- Preprint (755) (remove)
Language
- English (755)
Has Fulltext
- yes (755) (remove)
Is part of the Bibliography
- no (755) (remove)
Keywords
- Kongress (5)
- Kryptologie (5)
- Online-Publikation (4)
- Commitment Scheme (2)
- Oblivious Transfer (2)
- San Jose (2)
- computational complexity (2)
- Blind Signature (1)
- Block Korkin—Zolotarev reduction (1)
- Chinese Remainder Theorem (1)
Institute
- Informatik (755) (remove)
The human visual cortex enables visual perception through a cascade of hierarchical computations in cortical regions with distinct functionalities. Here, we introduce an AI-driven approach to discover the functional mapping of the visual cortex. We related human brain responses to scene images measured with functional MRI (fMRI) systematically to a diverse set of deep neural networks (DNNs) optimized to perform different scene perception tasks. We found a structured mapping between DNN tasks and brain regions along the ventral and dorsal visual streams. Low-level visual tasks mapped onto early brain regions, 3-dimensional scene perception tasks mapped onto the dorsal stream, and semantic tasks mapped onto the ventral stream. This mapping was of high fidelity, with more than 60% of the explainable variance in nine key regions being explained. Together, our results provide a novel functional mapping of the human visual cortex and demonstrate the power of the computational approach.
Dendritic spines are considered a morphological proxy for excitatory synapses, rendering them a target of many different lines of research. Over recent years, it has become possible to image simultaneously large numbers of dendritic spines in 3D volumes of neural tissue. In contrast, currently no automated method for spine detection exists that comes close to the detection performance reached by human experts. However, exploiting such datasets requires new tools for the fully automated detection and analysis of large numbers of spines. Here, we developed an efficient analysis pipeline to detect large numbers of dendritic spines in volumetric fluorescence imaging data. The core of our pipeline is a deep convolutional neural network, which was pretrained on a general-purpose image library, and then optimized on the spine detection task. This transfer learning approach is data efficient while achieving a high detection precision. To train and validate the model we generated a labelled dataset using five human expert annotators to account for the variability in human spine detection. The pipeline enables fully automated dendritic spine detection and reaches a near human-level detection performance. Our method for spine detection is fast, accurate and robust, and thus well suited for large-scale datasets with thousands of spines. The code is easily applicable to new datasets, achieving high detection performance, even without any retraining or adjustment of model parameters.
Integer point sets minimizing average pairwise L1 distance: What is the optimal shape of a town?
(2010)
An n-town, n[is an element of]N , is a group of n buildings, each occupying a distinct position on a 2-dimensional integer grid. If we measure the distance between two buildings along the axis-parallel street grid, then an n-town has optimal shape if the sum of all pairwise Manhattan distances is minimized. This problem has been studied for cities, i.e., the limiting case of very large n. For cities, it is known that the optimal shape can be described by a differential equation, for which no closed-form solution is known. We show that optimal n-towns can be computed in O(n[superscript 7.5]) time. This is also practically useful, as it allows us to compute optimal solutions up to n=80.
The dynamics of many systems are described by ordinary differential equations (ODE). Solving ODEs with standard methods (i.e. numerical integration) needs a high amount of computing time but only a small amount of storage memory. For some applications, e.g. short time weather forecast or real time robot control, long computation times are prohibitive. Is there a method which uses less computing time (but has drawbacks in other aspects, e.g. memory), so that the computation of ODEs gets faster? We will try to discuss this question for the assumption that the alternative computation method is a neural network which was trained on ODE dynamics and compare both methods using the same approximation error. This comparison is done with two different errors. First, we use the standard error that measures the difference between the approximation and the solution of the ODE which is hard to characterize. But in many cases, as for physics engines used in computer games, the shape of the approximation curve is important and not the exact values of the approximation. Therefore, we introduce a subjective error based on the Total Least Square Error (TLSE) which gives more consistent results. For the final performance comparison, we calculate the optimal resource usage for the neural network and evaluate it depending on the resolution of the interpolation points and the inter-point distance. Our conclusion gives a method to evaluate where neural nets are advantageous over numerical ODE integration and where this is not the case. Index Terms—ODE, neural nets, Euler method, approximation complexity, storage optimization.
At present, there are no quantitative, objective methods for diagnosing the Parkinson disease. Existing methods of quantitative analysis by myograms suffer by inaccuracy and patient strain; electronic tablet analysis is limited to the visible drawing, not including the writing forces and hand movements. In our paper we show how handwriting analysis can be obtained by a new electronic pen and new features of the recorded signals. This gives good results for diagnostics. Keywords: Parkinson diagnosis, electronic pen, automatic handwriting analysis
This paper describes the problems and an adaptive solution for process control in rubber industry. We show that the human and economical benefits of an adaptive solution for the approximation of process parameters are very attractive. The modeling of the industrial problem is done by the means of artificial neural networks. For the example of the extrusion of a rubber profile in tire production our method shows good results even using only a few training samples.
Pseudorandom function tribe ensembles based on one-way permutations: improvements and applications
(1999)
Pseudorandom function tribe ensembles are pseudorandom function ensembles that have an additional collision resistance property: almost all functions have disjoint ranges. We present an alternative to the construction of pseudorandom function tribe ensembles based on oneway permutations given by Canetti, Micciancio and Reingold [CMR98]. Our approach yields two different but related solutions: One construction is somewhat theoretic, but conceptually simple and therefore gives an easier proof that one-way permutations suffice to construct pseudorandom function tribe ensembles. The other, slightly more complicated solution provides a practical construction; it starts with an arbitrary pseudorandom function ensemble and assimilates the one-way permutation to this ensemble. Therefore, the second solution inherits important characteristics of the underlying pseudorandom function ensemble: it is almost as effcient and if the starting pseudorandom function ensemble is efficiently invertible (given the secret key) then so is the derived tribe ensemble. We also show that the latter solution yields so-called committing private-key encryption schemes. i.e., where each ciphertext corresponds to exactly one plaintext independently of the choice of the secret key or the random bits used in the encryption process.
We introduce the relationship between incremental cryptography and memory checkers. We present an incremental message authentication scheme based on the XOR MACs which supports insertion, deletion and other single block operations. Our scheme takes only a constant number of pseudorandom function evaluations for each update step and produces smaller authentication codes than the tree scheme presented in [BGG95]. Furthermore, it is secure against message substitution attacks, where the adversary is allowed to tamper messages before update steps, making it applicable to virus protection. From this scheme we derive memory checkers for data structures based on lists. Conversely, we use a lower bound for memory checkers to show that so-called message substitution detecting schemes produce signatures or authentication codes with size proportional to the message length.
We present efficient non-malleable commitment schemes based on standard assumptions such as RSA and Discrete-Log, and under the condition that the network provides publicly available RSA or Discrete-Log parameters generated by a trusted party. Our protocols require only three rounds and a few modular exponentiations. We also discuss the difference between the notion of non-malleable commitment schemes used by Dolev, Dwork and Naor [DDN00] and the one given by Di Crescenzo, Ishai and Ostrovsky [DIO98].
Based on the quadratic residuosity assumption we present a non-interactive crypto-computing protocol for the greater-than function, i.e., a non-interactive procedure between two parties such that only the relation of the parties' inputs is revealed. In comparison to previous solutions our protocol reduces the number of modular multiplications significantly. We also discuss applications to conditional oblivious transfer, private bidding and the millionaires' problem.
We propose a new security measure for commitment protocols, called Universally Composable (UC) Commitment. The measure guarantees that commitment protocols behave like an \ideal commitment service," even when concurrently composed with an arbitrary set of protocols. This is a strong guarantee: it implies that security is maintained even when an unbounded number of copies of the scheme are running concurrently, it implies non-malleability (not only with respect to other copies of the same protocol but even with respect to other protocols), it provides resilience to selective decommitment, and more. Unfortunately two-party uc commitment protocols do not exist in the plain model. However, we construct two-party uc commitment protocols, based on general complexity assumptions, in the common reference string model where all parties have access to a common string taken from a predetermined distribution. The protocols are non-interactive, in the sense that both the commitment and the opening phases consist of a single message from the committer to the receiver.
We review the representation problem based on factoring and show that this problem gives rise to alternative solutions to a lot of cryptographic protocols in the literature. And, while the solutions so far usually either rely on the RSA problem or the intractability of factoring integers of a special form (e.g., Blum integers), the solutions here work with the most general factoring assumption. Protocols we discuss include identification schemes secure against parallel attacks, secure signatures, blind signatures and (non-malleable) commitments.
We show that non-interactive statistically-secret bit commitment cannot be constructed from arbitrary black-box one-to-one trapdoor functions and thus from general public-key cryptosystems. Reducing the problems of non-interactive crypto-computing, rerandomizable encryption, and non-interactive statistically-sender-private oblivious transfer and low-communication private information retrieval to such commitment schemes, it follows that these primitives are neither constructible from one-to-one trapdoor functions and public-key encryption in general. Furthermore, our separation sheds some light on statistical zeroknowledge proofs. There is an oracle relative to which one-to-one trapdoor functions and one-way permutations exist, while the class of promise problems with statistical zero-knowledge proofs collapses in P. This indicates that nontrivial problems with statistical zero-knowledge proofs require more than (trapdoor) one-wayness.
We show lower bounds for the signature size of incremental schemes which are secure against substitution attacks and support single block replacement. We prove that for documents of n blocks such schemes produce signatures of \Omega(n^(1/(2+c))) bits for any constant c>0. For schemes accessing only a single block resp. a constant number of blocks for each replacement this bound can be raised to \Omega(n) resp. \Omega(sqrt(n)). Additionally, we show that our technique yields a new lower bound for memory checkers.
We introduce algorithms for lattice basis reduction that are improvements of the famous L3-algorithm. If a random L3-reduced lattice basis b1,b2,...,bn is given such that the vector of reduced Gram-Schmidt coefficients ({µi,j} 1<= j< i<= n) is uniformly distributed in [0,1)n(n-1)/2, then the pruned enumeration finds with positive probability a shortest lattice vector. We demonstrate the power of these algorithms by solving random subset sum problems of arbitrary density with 74 and 82 many weights, by breaking the Chor-Rivest cryptoscheme in dimensions 103 and 151 and by breaking Damgard's hash function.
We call a vector x/spl isin/R/sup n/ highly regular if it satisfies =0 for some short, non-zero integer vector m where <...> is the inner product. We present an algorithm which given x/spl isin/R/sup n/ and /spl alpha//spl isin/N finds a highly regular nearby point x' and a short integer relation m for x'. The nearby point x' is 'good' in the sense that no short relation m~ of length less than /spl alpha//2 exists for points x~ within half the x'-distance from x. The integer relation m for x' is for random x up to an average factor 2/sup /spl alpha//2/ a shortest integer relation for x'. Our algorithm uses, for arbitrary real input x, at most O(n/sup 4/(n+log A)) many arithmetical operations on real numbers. If a is rational the algorithm operates on integers having at most O(n/sup 5/+n/sup 3/(log /spl alpha/)/sup 2/+log(/spl par/qx/spl par//sup 2/)) many bits where q is the common denominator for x.
We report on improved practical algorithms for lattice basis reduction. We propose a practical floating point version of theL3-algorithm of Lenstra, Lenstra, Lovász (1982). We present a variant of theL3-algorithm with "deep insertions" and a practical algorithm for block Korkin—Zolotarev reduction, a concept introduced by Schnorr (1987). Empirical tests show that the strongest of these algorithms solves almost all subset sum problems with up to 66 random weights of arbitrary bit length within at most a few hours on a UNISYS 6000/70 or within a couple of minutes on a SPARC1 + computer.