Source author record

Arash Gholami Davoodi

Arash Gholami Davoodi appears in the imported research catalog. Authorship, coauthor and topic links are available while profile ownership is still unclaimed.

ResearcherUnclaimed source record

Catalog footprint

What is connected

6works
4topics
4close collaborators

Actions

Connect this record

Log in to claim

Research graph

See the researcher in context

Open full explorer

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

Published work

6 published item(s)

preprint2026arXiv

Vertex-Softmax: Tight Transformer Verification via Exact Softmax Optimization

Certified verification of transformer attention requires bounding the softmax function over interval constraints on the pre-softmax scores. Existing verifiers relax softmax ndependently of the downstream objective, leaving avoidable slack. We prove that the exact optimum of this score-box problem is attained at a vertex of the constraint box, and establish a threshold structure theorem showing that, after sorting the objective coefficients, the optimum lies among only linearly many candidates, yielding the Vertex-Softmax primitive with log-linear complexity in the sequence length. We further prove a formal optimality result showing that Vertex-Softmax is the tightest sound bound obtainable from score intervals alone, characterizing precisely what additional structure (score correlations, score-value coupling) is needed for further improvement. Integrated into a CROWN Convex Relaxation based Optimization for Worst-case Neurons)-style verifier with a formal soundness guarantee, Vertex-Softmax significantly improves certified rates and substantially tightens lower bounds across MNIST, Fashion-MNIST, and CIFAR-10 attention models, while consistently matching or outperforming alpha-CROWN and branch-and-bound baselines at a fraction of their cost.

preprint2020arXiv

ForestDSH: A Universal Hash Design for Discrete Probability Distributions

In this paper, we consider the problem of classification of $M$ high dimensional queries $y^1,\cdots,y^M\in B^S$ to $N$ high dimensional classes $x^1,\cdots,x^N\in A^S$ where $A$ and $B$ are discrete alphabets and the probabilistic model that relates data to the classes $P(x,y)$ is known. This problem has applications in various fields including the database search problem in mass spectrometry. The problem is analogous to the nearest neighbor search problem, where the goal is to find the data point in a database that is the most similar to a query point. The state of the art method for solving an approximate version of the nearest neighbor search problem in high dimensions is locality sensitive hashing (LSH). LSH is based on designing hash functions that map near points to the same buckets with a probability higher than random (far) points. To solve our high dimensional classification problem, we introduce distribution sensitive hashes that map jointly generated pairs $(x,y)\sim P$ to the same bucket with probability higher than random pairs $x\sim P^A$ and $y\sim P^B$, where $P^A$ and $P^B$ are the marginal probability distributions of $P$. We design distribution sensitive hashes using a forest of decision trees and we show that the complexity of search grows with $O(N^{λ^*(P)})$ where $λ^*(P)$ is expressed in an analytical form. We further show that the proposed hashes perform faster than state of the art approximate nearest neighbor search methods for a range of probability distributions, in both theory and simulations. Finally, we apply our method to the spectral library search problem in mass spectrometry, and show that it is an order of magnitude faster than the state of the art methods.

preprint2016arXiv

GDoF of the MISO BC: Bridging the Gap between Finite Precision and Perfect CSIT

For the $K=2$ user MISO BC, i.e., the wireless broadcast channel where a transmitter equipped with $K=2$ antennas sends independent messages to $K=2$ receivers each of which is equipped with a single antenna, the sum generalized degrees of freedom (GDoF) are characterized for arbitrary channel strength and channel uncertainty levels for each of the channel coefficients. The result is extended to $K>2$ users under additional restrictions which include the assumption of symmetry.

preprint2016arXiv

Generalized Degrees of Freedom of the Symmetric K-User Interference Channel under Finite Precision CSIT

The generalized degrees of freedom (GDoF) characterization of the symmetric K-user interference channel is obtained under finite precision channel state information at the transmitters (CSIT). The symmetric setting is where each cross channel is capable of carrying degrees of freedom (DoF) while each direct channel is capable of carrying 1 DoF. Remarkably, under finite precision CSIT the symmetric K-user interference channel loses all the GDoF benefits of interference alignment. The GDoF per user diminish with the number of users everywhere except in the very strong (optimal for every receiver to decode all messages) and very weak (optimal to treat all interference as noise) interference regimes. The result stands in sharp contrast to prior work on the symmetric setting under perfect CSIT, where the GDoF per user remain undiminished due to interference alignment. The result also stands in contrast to prior work on a subclass of asymmetric settings under finite precision CSIT, i.e., the topological interference management problem, where interference alignment plays a crucial role and provides substantial GDoF benefits.

preprint2014arXiv

Aligned Image Sets under Channel Uncertainty: Settling a Conjecture by Lapidoth, Shamai and Wigger on the Collapse of Degrees of Freedom under Finite Precision CSIT

A conjecture made by Lapidoth, Shamai and Wigger at Allerton 2005 (also an open problem presented at ITA 2006) states that the DoF of a 2 user broadcast channel, where the transmitter is equipped with 2 antennas and each user is equipped with 1 antenna, must collapse under finite precision CSIT. In this work we prove that the conjecture is true in all non-degenerate settings (e.g., where the probability density function of unknown channel coefficients exists and is bounded). The DoF collapse even when perfect channel knowledge for one user is available to the transmitter. This also settles a related recent conjecture by Tandon et al. The key to our proof is a bound on the number of codewords that can cast the same image (within noise distortion) at the undesired receiver whose channel is subject to finite precision CSIT, while remaining resolvable at the desired receiver whose channel is precisely known by the transmitter. We are also able to generalize the result along two directions. First, if the peak of the probability density function is allowed to scale as O(P^(α/2)), representing the concentration of probability density (improving CSIT) due to, e.g., quantized feedback at rate (α/2)\log(P), then the DoF are bounded above by 1+α, which is also achievable under quantized feedback. Second, we generalize the result to the K user broadcast channel with K antennas at the transmitter and a single antenna at each receiver. Here also the DoF collapse under non-degenerate channel uncertainty. The result directly implies a collapse of DoF to unity under non-degenerate channel uncertainty for the general K-user interference and MxN user X networks as well.

preprint2012arXiv

Optimum Power Allocations for Fading Decode-and-Forward Relay Channel

In this paper, for a fading decode-and-forward full-duplex relay channel, we analytically derive optimum power allocations. Individual power constraints for the source and the relay are assumed and the related optimization problem is analyzed for two scenarios. First, optimization is taken over the source power, the relay power, and the correlation coefficient between the transmitted signals of the source and the relay. Then, for a fixed value of correlation coefficient, the optimization problem is analyzed. It is also proven that the optimization problems are convex for these two scenarios. Finally, implications of theoretical results are discussed through simulations for each scenario.