no code implementations • 8 Feb 2024 • Amir Zandieh, Insu Han, Vahab Mirrokni, Amin Karbasi
In this work, our focus is on developing an efficient compression technique for the KV cache.
1 code implementation • 9 Oct 2023 • Insu Han, Rajesh Jayaram, Amin Karbasi, Vahab Mirrokni, David P. Woodruff, Amir Zandieh
Recent work suggests that in the worst-case scenario, quadratic time is necessary unless the entries of the attention matrix are bounded or the matrix has low stable rank.
1 code implementation • 5 Feb 2023 • Amir Zandieh, Insu Han, Majid Daliri, Amin Karbasi
Dot-product attention mechanism plays a crucial role in modern deep architectures (e. g., Transformer) for sequence modeling, however, na\"ive exact computation of this model incurs quadratic time and memory complexities in sequence length, hindering the training of long-sequence models.
2 code implementations • 9 Sep 2022 • Insu Han, Amir Zandieh, Jaehoon Lee, Roman Novak, Lechao Xiao, Amin Karbasi
Moreover, most prior works on neural kernels have focused on the ReLU activation, mainly due to its popularity but also due to the difficulty of computing such kernels for general activations.
no code implementations • 9 Feb 2022 • David P. Woodruff, Amir Zandieh
We propose an input sparsity time sampling algorithm that can spectrally approximate the Gram matrix corresponding to the $q$-fold column-wise tensor product of $q$ matrices using a nearly optimal number of samples, improving upon all previously known methods by poly$(q)$ factors.
no code implementations • 7 Feb 2022 • Insu Han, Amir Zandieh, Haim Avron
Our proposed GZK family, generalizes the zonal kernels (i. e., dot-product kernels on the unit sphere) by introducing radial factors in their Gegenbauer series expansion, and includes a wide range of ubiquitous kernel functions such as the entirety of dot-product kernels as well as the Gaussian and the recently introduced Neural Tangent kernels.
1 code implementation • NeurIPS 2021 • Amir Zandieh, Insu Han, Haim Avron, Neta Shoham, Chaewon Kim, Jinwoo Shin
To accelerate learning with NTK, we design a near input-sparsity time approximation algorithm for NTK, by sketching the polynomial expansions of arc-cosine kernels: our sketch for the convolutional counterpart of NTK (CNTK) can transform any image using a linear runtime in the number of pixels.
no code implementations • 1 Apr 2021 • Amir Zandieh
The Neural Tangent Kernel (NTK) characterizes the behavior of infinitely wide neural nets trained under least squares loss by gradient descent.
no code implementations • 21 Mar 2020 • Michael Kapralov, Navid Nouri, Ilya Razenshteyn, Ameya Velingker, Amir Zandieh
Random binning features, introduced in the seminal paper of Rahimi and Recht (2007), are an efficient method for approximating a kernel matrix using locality sensitive hashing.
1 code implementation • NeurIPS 2019 • Andisheh Amrollahi, Amir Zandieh, Michael Kapralov, Andreas Krause
In this paper we consider the problem of efficiently learning set functions that are defined over a ground set of size $n$ and that are sparse (say $k$-sparse) in the Fourier domain.
1 code implementation • 3 Sep 2019 • Thomas D. Ahle, Michael Kapralov, Jakob B. T. Knudsen, Rasmus Pagh, Ameya Velingker, David Woodruff, Amir Zandieh
Oblivious sketching has emerged as a powerful approach to speeding up numerical linear algebra over the past decade, but our understanding of oblivious sketching solutions for kernel matrices has remained quite limited, suffering from the aforementioned exponential dependence on input parameters.
Data Structures and Algorithms
no code implementations • 20 Dec 2018 • Haim Avron, Michael Kapralov, Cameron Musco, Christopher Musco, Ameya Velingker, Amir Zandieh
We formalize this intuition by showing that, roughly, a continuous signal from a given class can be approximately reconstructed using a number of samples proportional to the *statistical dimension* of the allowed power spectrum of that class.
no code implementations • 6 Aug 2018 • Ashkan Norouzi-Fard, Jakub Tarnawski, Slobodan Mitrović, Amir Zandieh, Aida Mousavifar, Ola Svensson
It is the first low-memory, single-pass algorithm that improves the factor $0. 5$, under the natural assumption that elements arrive in a random order.
no code implementations • ICML 2018 • Ashkan Norouzi-Fard, Jakub Tarnawski, Slobodan Mitrovic, Amir Zandieh, Aidasadat Mousavifar, Ola Svensson
It is the first low-memory, singlepass algorithm that improves the factor 0. 5, under the natural assumption that elements arrive in a random order.
no code implementations • ICML 2017 • Haim Avron, Michael Kapralov, Cameron Musco, Christopher Musco, Ameya Velingker, Amir Zandieh
Qualitatively, our results are twofold: on the one hand, we show that random Fourier feature approximation can provably speed up kernel ridge regression under reasonable assumptions.