Training DNNs with Hybrid Block Floating Point

Chat with Graph Search

Ask any question about EPFL courses, lectures, exercises, research, news, etc. or try the example questions below.

DISCLAIMER: The Graph Chatbot is not programmed to provide explicit or categorical answers to your questions. Rather, it transforms your questions into API requests that are distributed across the various IT services officially administered by EPFL. Its purpose is solely to collect and recommend relevant references to content that you can explore to help you answer your questions.

Smart contracts have emerged as the most promising foundations for applications of the blockchain technology. Even though smart contracts are expected to serve as the backbone of the next-generation web, they have several limitations that hinder their wide ...

Situational awareness strategies are essential for the reliable and secure operation of the electric power grid which represents critical infrastructure in modern society. With the rise of converter-interfaced renewable generation and the consequent shift ...

State-of-the-art Artificial Intelligence (AI) algorithms, such as graph neural networks and recommendation systems, require floating-point computation of very large matrix multiplications over sparse data. Their execution in resource-constrained scenarios, ...

Compute memories are memory arrays augmented with dedicated logic to support arithmetic. They support the efficient execution of data-centric computing patterns, such as those characterizing Artificial Intelligence (AI) algorithms. These architectures can ...

Hand gestures are one of the most natural and expressive way for humans to convey information, and thus hand gesture recognition has become a research hotspot in the human-machine interface (HMI) field. In particular, biological signals such as surface ele ...

We use numerical bootstrap techniques to study correlation functions of traceless sym-metric tensors of O(N) with two indices ti j. We obtain upper bounds on operator dimen-sions for all the relevant representations and several values of N. We discover sev ...

E. E. Floyd showed in 1973 that there exist only two nontrivial cobor-dism classes that contain manifolds with three cells, and that they lie in dimen-sions 10 and 5. We prove that there is an action of the cyclic group C2 on the 10-dimensional Floyd manif ...

In this thesis, we unveil a third design path to manipulate elastic waves within architected media, distinct from the traditional phononic crystal and locally-resonant metamaterial concepts. The core innovation lies in the concept of nonlocal resonances, d ...

The high computational costs of deep convolutional neural networks hinder their deployment in real-world applications, including pulmonary nodule detection from CT scans where large 3D image sizes amplify the issue. This paper presents a novel 3D method to ...

Correlated errors of experimental data are a common but often neglected problem in physical sciences. Various tools are provided here for thorough propagation of uncertainties in cases of correlated errors. Discussed are techniques especially applicable to ...

With the recent advent of blockchains, we have witnessed a plethora of blockchain proposals. These proposals range from using work to using time, storage or stake in order to select blocks to be appended to the chain. As a drawback it makes it difficult fo ...

Effectively localizing an agent in a realistic, noisy setting is crucial for many embodied vision tasks. Visual Odometry (VO) is a practical substitute for unreliable GPS and compass sensors, especially in indoor environments. While SLAM-based methods show ...

The study of non-contact manipulation in water, and the ability to robotically control floating objects has gained recent attention due to wide-ranging potential applications, including the analysis of plastic pollution in the oceans and the optimization o ...

The Supersingular Isogeny Diffie-Hellman (SIDH) protocol has been the main and most efficient isogeny-based encryption protocol, until a series of breakthroughs led to a polynomial-time key-recovery attack. While some countermeasures have been proposed, th ...

We propose a method for adapting neural networks to distribution shifts at test-time. In contrast to training-time robustness mechanisms that attempt to anticipate and counter the shift, we create a closed-loop system and make use of test-time feedback sig ...

We show the Jordan property for regional fundamental groups of klt singularities of fixed dimension. Furthermore, we prove the existence of effective simultaneous index 1 covers for n-dimensional klt singularities. We give an application to the study of lo ...

We propose nonparametric estimators for the second-order central moments of possibly anisotropic spherical random fields, within a functional data analysis context. We consider a measurement framework where each random field among an identically distribute ...

State-of-the-art training algorithms for deep learning models are based on stochastic gradient descent (SGD). Recently, many variations have been explored: perturbing parameters for better accuracy (such as in Extra-gradient), limiting SGD updates to a sub ...

Shadows for bicategories, defined by Ponto, provide a useful framework that generalizes classical and topological Hochschild homology. In this paper, we define Hochschild-type invariants for monoids in a symmetric monoidal, simplicial model category V, as ...

LES and wind tunnel studies have shown significant benefit when allowing turbines (T) in a wind farm to adopt different heights. This work presents two new genetic algorithms (GA) that perform wind farm layout optimization (WFLO) involving continuous and t ...

Training DNNs with Hybrid Block Floating Point

Chat with Graph Search

Towards General-Purpose Decentralized Computing with Permissionless Extensibility

Functional-Basis Analysis of Non-Stationary Signals in Modern Power Grids: Theory and Implementation in Embedded Systems

A 16-bit Floating-Point Near-SRAM Architecture for Low-power Sparse Matrix-Vector Multiplication

Overflow-free compute memories for edge AI acceleration

Motor-Unit Ordering of Blindly-Separated Surface-EMG Signals for Gesture Recognition

Bootstrapping traceless symmetric O(N) scalars

Floyd's manifold is a conjugation space

Nonlocally-resonant metamaterials

Efficient Lung Nodule Detection via 3D Deep Learning with Shifted Convolutions

Reliable uncertainties: Error correlation, rotated error bars, and linear regressions in three-isotope plots and beyond

DIABLO: A Benchmark Suite for Blockchains

Modality-invariant Visual Odometry for Embodied Vision

Non-contact robotic manipulation of floating objects: exploiting emergent limit cycles

New SIDH Countermeasures for a More Efficient Key Exchange

Rapid Network Adaptation: Learning to Adapt Neural Networks Using Test-Time Feedback

The Jordan property for local fundamental groups

Functional estimation of anisotropic covariance and autocovariance operators on the sphere

Masked Training of Neural Networks with Partial Gradients

A Shadow Perspective on Equivariant Hochschild Homologies

Wind farm layout and unconstrained hub height optimization using genetic algorithms applied to different power densities