Dark Web ChatGPT Unleashed: Meet DarkBERT

 Training on the Dark Web?

Dark Web ChatGPT Unleashed: Meet DarkBERT      in CACM   By Tom's Hardware, May 19, 2023

To train the model, the researchers crawled the Dark Web through the Tor network, then filtered the raw data (applying techniques such as deduplication, category balancing, and data pre-processing) to generate a Dark Web database.

Researchers at South Korea's Korea Advanced Institute of Science and Technology (KAIST) and data intelligence company S2W have created a large language model (LLM) trained on Dark Web data.

The researchers fed the RoBERTa framework a database they compiled from the Dark Web via the Tor network to create the DarkBERT LLM, which can analyze and extract useful information from a new piece of Dark Web content composed in its own dialects and heavily-coded messages.

They demonstrated DarkBERT's superior performance to other LLMs, which should enable security researchers and law enforcement to delve deeper into the Dark Web.

From Tom's Hardware

View Full Article   

Bacteria Fingerprint ID

 Identifying bacteria using Machine Learning.

'Fingerprint' ML Technique Identifies Bacteria in Seconds, KAIST (South Korea), March 4, 2022

Researchers at the Korea Advanced Institute of Science and Technology (KAIST) combined surface-enhanced Raman spectroscopy and a deep learning model to identify bacteria in seconds with up to 98% accuracy. Their model, named DualWKNet (dual-branch wide-kernel network), was trained to identify the "fingerprint" spectra of the molecular components of multiple bacteria. Said KAIST's Sungho Jo, "We demonstrated a markedly simple, fast, and effective route to classify the signals of two common bacteria and their resident media without any separation procedures." Jo added, "Ultimately, with the use of DualWKNet replacing the bacteria and media separation steps, our method dramatically reduces analysis time."

Detecting, Recognizing Voices at a Distance

Hmm, talk about loss of privacy.   Will this do it through a mask? Will we be wearing new kinds of 'voice' masks?  Note the training data.  Probably will be quickly restricted for use in the West.   But elsewhere?  

Biomimetic Resonant Acoustic Sensor Detecting Far-Distant Voices Accurately to Hit the Market

KAIST (South Korea), June 14, 2021  in CACM

Researchers at South Korea's Korea Advanced Institute of Science and Technology (KAIST) have developed a bioinspired flexible piezoelectric acoustic sensor with a multi-resonant ultrathin piezoelectric membrane that acts like the basilar membrane of the human cochlea to achieve accurate and far-distant voice detection. The miniaturized sensor can be embedded into smartphones and artificial intelligence speakers for machine learning-based biometric authentication and voice processing. Compared to a MEMS condenser microphone, the researchers found the speaker identification error rate for their resonant mobile acoustic sensor was 56% lower after it experienced 150 training datasets, and 75% lower after 2,800 training datasets. ... ' 

Exploring Interactions with Haptic Feedback in Virtual Reality

Will this mean that people will be able to immerse themselves in games and imulation?   I am not much of a gamer, but like the idea that people will more realistic  'digital twins' to engage with physical objects and spaces.   Consider the future of that.  Its not only game-like controllers, but 'immersive interactions' that can enable us to be part of our physically enabled world.   Inside a 'digital twin'?  A powerful illusion indeed 

Microsoft Research collaborates with KAIST in Korea to explore bimanual interactions with haptic feedback in virtual reality

Published May 6, 2021

By Michel Pahud , Principal Research Software Development Engineer  Mike Sinclair , Senior Principal Researcher  Andrea Bianchi , Associate Professor at KAIST

Editor’s Note: Bimanual controllers are frequently used to enhance the realism and immersion of virtual reality experiences such as games and simulations. Researchers have typically relied on mechanical linkages between the controllers to recreate the sensation of holding different objects with both hands. However, those linkages cannot quickly adapt to simulate dynamic objects. They also make for bulky controllers that can’t be disconnected to support free, independent movements. This is the problem that researchers seek to solve in the recent paper titled “GamesBond: Bimanual Haptic Illusion of Physically Connected Objects for Immersive VR Using Grip Deformation”.

GamesBond is the outcome of a recent collaboration between Michel Pahud and Mike Sinclair from Microsoft Research and Andrea Bianchi, associate professor at KAIST and director of the MAKinteract lab, with two of his students, Neung Ryu, the original author of the paper, and Hye-Young Jo. The paper was accepted at ACM CHI Conference on Human Factors in Computing Systems (CHI 2021), where it received an honorable mention award.

In this project, we explored a pair of novel 4-DoF controllers, without actual physical linkage between them, that can bend, twist, and stretch together in concert to create the illusion of being connected as a single device with a physical link. Each controller can bend from 0 to 30 degrees in any direction, twist from -70 degrees to 70 degrees and stretch from -2.5mm to 9.0mm (the paper provides all the details of the mechanism).  ... " 

Knock it with a SmartPhone to Identify

Here a system built in South Korea that lets you knock on an object with a smartphone, and it identifies it using sound and motion sensors.   We worked on problems that predicted maintenance needs of machinery, using sounds and vibrations, might this be useful for such an application as well?  Another way to gather data that might be useful otherwise.     Add visual inspection information for machinery diagnosis?

Object Identification, Interaction with a Smartphone Knock
KAIST - South Korea

Researchers at the Korea Advanced Institute of Science and Technology (KAIST) in South Korea have developed technology for identifying objects and carrying out actions, simply by knocking on objects with a smartphone. The Knocker solution combines the sound and motion sensors in smartphones with software that executes directions according to sounds and vibrations. These capabilities are maintained even in low-light settings, without the use of cameras or other specialized hardware. Knocker harnesses the smartphone's built-in sensors, and machine learning identifies objects the device is knocked on by analyzing the responses generated when a smartphone is knocked against an object. Knocker was able to identify 23 everyday objects in noisy environments with 83% accuracy, and with 98% accuracy in quiet indoor environments.  ... "