Wednesday, June 28, 2023

Researchers’ warning on childhood vaccination target

A third of under-fives attending a Paediatric Emergency Department (PED) and who are eligible for pre-school boosters have unmet vaccination need according to research from Lancaster Medical School

Peer-Reviewed Publication

LANCASTER UNIVERSITY

A third of under-fives attending a Paediatric Emergency Department (PED) and who are eligible for pre-school boosters have unmet vaccination need according to research from Lancaster Medical School.

The study published in BMJ Open examined the percentage of children aged between two months and five years who were up-to-date with their routine childhood vaccinations at their time of attendance at the Paediatric Emergency Department (PED) at North Manchester General Hospital (NMGH) – part of Manchester University NHS Foundation Trust (MFT) - in October 2021.

Children and young people in Manchester have lower than average levels of health and wellbeing, around a third live in relative low-income households, and 1% are in care.

Looking at tetanus and measles, mumps, and rubella (MMR) vaccination (as a proxy for overall vaccination status), researchers found that a third of under-fives were missing at least one dose.

The study was led by Professor Rachel Isba who is a Professor of Medicine at Lancaster Medical School, an NHS Consultant in Paediatric Public Health Medicine for Alder Hey Children’s Hospital in Liverpool, and an Honorary Consultant in Paediatric Public Health Medicine at Royal Manchester Children’s Hospital – also part of MFT.

The study found that in older age groups, many children were missing their tetanus boosters and only 1 in 5 of those eligible had received two doses of MMR. Those in younger age groups had vaccination coverage levels comparable to the local data, but still below the target of 95%.

Professor Isba said: “Vaccine-preventable diseases can be very serious and measles cases are on the increase. Those children eligible for pre-school boosters - tetanus and MMR2 - appear to have considerable unmet vaccination need. A hospital attendance provides an opportunity to check the vaccination status of children and young people.”

William Roberts, CEO of the Royal Society for Public Health, said : "This really important piece of research highlights that we must use every opportunity to support children and families to access vaccinations so that we can reach the coverage levels needed to protect communities. Children and young people have told RSPH that vaccines are important to their health, and so it is up to us to act collectively, ensuring children and families from all backgrounds can have their vaccination needs met. We know that people have vaccines if they are easily accessible and that is why initiatives like "Grab-a-jab" are so successful at reaching a wide range of communities."

Professor Isba said: “There are pockets of children with very low MMR vaccine coverage who are therefore at high risk of illness. Whilst the pandemic has had an impact, the observation that MMR2 uptake is considerably lower than tetanus booster when they are scheduled together warrants further investigation.

“Catch-up campaigns for MMR should include a focus on this cohort of children (who are now around six years old) and attendance at the Paediatric Emergency Department may offer an opportunity for us to offer an intervention.”

Dr Helen Wall, Clinical Director for Population Health, NHS Greater Manchester Integrated Care, said: “Measles, mumps and rubella are highly infectious diseases. They can make children extremely ill and lead to health complications such as deafness, fits and brain damage. Vaccinations are thoroughly tested and are one of the most important things we can do to protect ourselves, our children and those around us against ill health.

“It is never too late to catch up on any missed doses of MMR vaccination. You can contact your GP to check whether your child is up to date with their vaccinations and book an appointment.”

JOURNAL

BMJ Open

DOI

10.1136/bmjopen-2023-072053

METHOD OF RESEARCH

Data/statistical analysis

SUBJECT OF RESEARCH

People

ARTICLE TITLE

Unmet vaccination need among children under the age of five attending the paediatric emergency department: a cross-sectional study in a large UK district general hospital

ARTICLE PUBLICATION DATE

27-Jun-2023

Keeping your backdoor secure—in your robust machine learning model

SUTD researchers developed AEGIS, the first and only technique that is able to detect backdoor attacks in robust machine learning models. Analysing robust models with AEGIS would improve the overall trustworthiness of artificial intelligence

Peer-Reviewed Publication

SINGAPORE UNIVERSITY OF TECHNOLOGY AND DESIGN

Fig. 1: An example of a typical backdoor attack (adapted from Wang et al. (2019) — IMAGE: THE VISIBLE DISTRIBUTED TRIGGER IS SHOWN IN FIGURE 1(A) AND THE TARGET LABEL IS SEVEN (7). THE TRAINING DATA IS MODIFIED. WE SEE THIS IN FIGURE 1(B) AND THE MODEL IS TRAINED WITH THIS POISONED DATA. THE INPUTS WITHOUT THE TRIGGER WILL BE CORRECTLY CLASSIFIED AND THE ONES WITH THE TRIGGER WILL BE INCORRECTLY CLASSIFIED DURING THE INFERENCE, AS SEEN IN FIGURE 1(C). view more
CREDIT: SUTD

Software systems are all around us—from the operating system of our computers to search engines to automation used in industrial applications. At the centre of all of this is data, which is used in machine learning (ML) components that are available in a wide variety of applications, including self-driving cars and large language models (LLM). Because many systems rely on ML components, it is important to guarantee their security and reliability.

For ML models trained using robust optimisation methods (that is, robust ML models), their effectiveness against various attacks is unknown. An example of a major attack vector is backdoor poisoning, which refers to compromised training data fed into the model. Technologies that detect backdoor attacks in standard ML models exist, but robust models require different detection methods for backdoor attacks because they behave differently than standard models and hold different assumptions.

This is the gap that Dr Sudipta Chattopadhyay, Assistant Professor at the Information Systems Technology and Design (ISTD) Pillar of the Singapore University of Technology and Design (SUTD), aimed to close.

In the study ‘Towards backdoor attacks and defense in robust machine learning models’, published in Computers & Security, Asst Prof Chattopadhyay and fellow SUTD researchers studied how to inject and defend against backdoor attacks for robust models in a certain ML component called image classifiers. Specifically, the models studied were trained using the state-of-the-art projected gradient descent (PGD) method.

The backdoor issue is urgent and dangerous, especially because of how current software pipelines are developed. Asst Prof Chattopadhyay stated, “No one develops a ML model pipeline and data collection from scratch nowadays. They might download training data from the internet or even use a pre-trained model. If the pre-trained model or dataset is poisoned, the resulting software, using these models, will be insecure. Often, only 1% of data poisoning is needed to create a backdoor.”

The difficulty with backdoor attacks is that only the attacker knows the pattern of poisoning. The user cannot go through this poison pattern to recognize whether their ML model has been infected. “The difficulty of the problem fascinated us. We speculated that the internals of a backdoor model might be different than a clean model,” said Asst Prof Chattopadhyay.

To this end, Asst Prof Chattopadhyay investigated backdoor attacks for robust models and found that they are highly susceptible (67.8% success rate). He also found that poisoning a training set creates mixed input distributions for the poisoned class, enabling the robust model to learn multiple feature representations for a certain prediction class. In contrast, clean models will only learn a single feature representation for a certain prediction class.

Along with fellow researchers, Asst Prof Chattopadhyay used this fact to his advantage to develop AEGIS, the very first backdoor detection technique for PGD-trained robust models. Using t-Distributed Stochastic Neighbour Embedding (t-SNE) and Mean Shift Clustering as a dimensionality reduction technique and clustering method, respectively, AEGIS is able to detect multiple feature representations in a class and identify backdoor-infected models.

AEGIS operates in five steps—it (1) uses an algorithm to generate translated images, (2) extracts feature representations from the clean training and clean/backdoored translated images, (3) reduces the dimensions of the extracted features via t-SNE, (4) employs mean shift to calculate the clusters of the reduced feature representations, and (5) counts these clusters to determine if the model is backdoor-infected or clean.

If there are two clusters (the training images and the translated images) in a model, then AEGIS flags this model as clean. If there are more than two clusters (the training images, the clean translated images, and the poisoned translated images), then AEGIS flags this model as suspicious and backdoor-infected.

Further, AEGIS effectively detected 91.6% of all backdoor-infected robust models with only a false positive rate of 11.1%, showing its high efficacy. As even the top backdoor detection technique in standard models is unable to flag backdoors in robust models, the development of AEGIS is important. It is critical to note that AEGIS is specialised to detect backdoor attacks in robust models and is ineffective in standard models.

Besides the ability to detect backdoor attacks in robust models, AEGIS is also efficient. Compared to standard backdoor defences that take hours to days to identify a backdoor-infected model, AEGIS only takes an average of five to nine minutes. In the future, Asst Prof Chattopadhyay aims to further refine AEGIS so that it can work with different and more complicated data distributions to defend against more threat models besides backdoor attacks.

Acknowledging the buzz around artificial intelligence (AI) in today’s climate, Asst Prof Chattopadhyay expressed, “We hope that people are aware of the risks associated with AI. Technologies powered by LLM like ChatGPT are trending, but there are huge risks and backdoor attacks are just one of them. With our research, we aim to achieve the adoption of trustworthy AI.”

An attack Model for AEGIS.

CREDIT

SUTD