Computer Vision

BC-Z aims to offer "zero-shot" learning for generalized autonomous robotics

Using a now public-access dataset, a research team has created a robotics control system which can generalize to unseen related tasks — allowing robots to interpret natural-language commands and video demonstrations.

Autonomous drones, triggered by a machine learning algorithm, could solve the urban feral pigeon problem

With research suggesting feral pigeons do as much as $1.1 billion in damage in the US alone, a project which pairs a commercial drone with machine learning to scare them away — without injury — could prove key to their control.

Did you check for potential errors in the labeled training dataset?

Attention: Don’t presume human or model-generated labels to be the growth truth for machine learning deployment in autonomous vehicular systems.

Can Retinal Scans Help in Predicting Myocardial Infarction?

The proposed AI system has an accuracy between 70% and 80% that can be employed as a second referral mechanism for cardiovascular examination.

Data Annotation for Computer Vision

5 Questions to Ask Before Getting Started with Data Annotation: What is data annotation? Why is it essential? What are common label types for models? How do AI-powered data annotation tools help with the computer vision labeling process? What’s the best way to get started?

Sage Elliott

 from 

Plainsight

-

Updated on 10 Feb, 2022

4 minutes read

Reconfigured AI devices enable lifelong learning for machines

By reconfiguring neural networks in artificial intelligence (AI) devices, a multi-institute team that included Penn State researchers facilitated AI systems to continually learn and adapt new data and tasks in ways that were not possible or practical before.

Sarah Small

 from 

Penn State

-

Updated on 10 Feb, 2022

4 minutes read

CyberSpec turns machine learning onto the problem of spectrum sensor attacks, without overloading the host

The CyberSpec framework is designed to detect anomalous behavior linked to cyber-attacks against crowd-sensing spectrum sensors, even when said sensors are running on lightweight resource-constrained hardware like a Raspberry Pi.

HyperTransformer bests rivals at few-shot learning - and takes advantage of unlabeled samples

Developed at Google Research, HyperTransformer decouples the task space and individual task complexity to generate all model weights in just one pass — while also offering support for unlabeled sample ingestion.

One-short learning approach, inspired by human intuition, works with just one training image

Designed to dramatically reduce the amount of training data needed for an image recognition system, this one-shot approach "inspired by nativism" takes a leaf from humans' ability to intuit and abstract.

From Embedded Sensors to Advanced Intelligence: Driving Industry 4.0 Innovation with TinyML

ML can be deployed on Embedded Sensors to unleash the full power of Industry 4.0.

Researchers program an autonomous vehicle to drift around obstacles using Nonlinear Model Predictive Control (NMPC)

Advances in software allow a customized car to perform controlled, autonomous drifting to enhance active safety and to give drivers the skills of professional racers.

Advanced Split Computing Strategy to Achieve High Accuracy under Strong Compression Rates

BottleFit framework applied DNN models in image classification achieves 77.1% data compression with up to 0.6% accuracy loss on the ImageNet dataset.

Applications of TinyML making specialised education more accessible

In this article, we look at two tinyML projects for education. We show how Backyard Brains uses low-cost experiment kits to make neuroscience education more accessible. We also introduce our readers to a specialisation offered by Harvard & Google to help students learn tinyML like never before.

Researchers point to pure-convolutional networks as a key for next-generation computer vision

Considered obsolete since the introduction of vision transformers, ConvNeXt proves there's life in convolution yet — outperforming its rivals by adopting some of their own tricks.

Meta AI Introduces Data2vec Open-Source Framework for Self-Supervised Learning in NLP, Speech and Vision Processing

A holistic open-source framework to improve the performance of multiple modalities and outperform the existing single-modality algorithms.