A member of technical staff at @Ataraxis_AI
Joined November 2012
- Tweets191
- Following265
- Followers179
- Likes1.2K
Jungkyu Park retweeted
We've recently developed new causal AI methods that allow us to simulate patient outcomes under counterfactual treatment scenarios
Importantly, as we start to scale these world models, they begin to generalize zero-shot to multiple solid tumors - incredible work by our team:
Today, Ataraxis is announcing a new generation of AI models that learn to predict treatment response and patient outcomes, and generalize zero-shot from breast cancer to multiple solid tumors.
Blog post: ataraxis.ai/causal-ai-chemot…
Jungkyu Park retweeted
Today, Ataraxis is announcing a new generation of AI models that learn to predict treatment response and patient outcomes, and generalize zero-shot from breast cancer to multiple solid tumors.
Blog post: ataraxis.ai/causal-ai-chemot…
Jungkyu Park retweeted
How aggressive a cancer is and how large is the benefit from chemotherapy are different estimands. Genomic assays were built decades ago for the first and repurposed for the second.
Today, we released 3 preprints on estimating the benefit of escalated treatment directly.
Jungkyu Park retweeted
What if a routine breast cancer H&E slide could help answer a hard question: how likely is this cancer to come back? Our @NatureComms paper introduces a multimodal AI test for recurrence risk across invasive breast cancer subtypes.
nature.com/articles/s41467-0…
1/
Jungkyu Park retweeted
We are looking for exceptional candidates for technical roles at @Ataraxis_AI:
* Research Engineer (Optimization/Systems): jobs.ashbyhq.com/ataraxis-ai…
* Research Engineer (Data Science): jobs.ashbyhq.com/ataraxis-ai…
* Research Scientist (Causality): jobs.ashbyhq.com/ataraxis-ai…
Join us!
Jungkyu Park retweeted
Today, we announce a strategic research collaboration between Ataraxis AI and MEDSIR (@wearemedsir), a global leader in oncology research.
The partnership will evaluate the Ataraxis Breast platform in several international randomized controlled trials. Together, we aim to generate robust clinical evidence on the role of artificial intelligence in predicting outcomes and treatment effects in early-stage and metastatic breast cancer. The analyses will especially focus on novel therapies investigated in MEDSIR's groundbreaking trials, including CDK4/6 inhibitors and antibody-drug conjugates (ADCs).
This is one of many ongoing partnerships at Ataraxis working on developing a robust evidence base supporting the accuracy and utility of Ataraxis AI-native tools, and moving AI precision oncology closer to becoming the standard of care.
Read the full announcement here: businesswire.com/news/home/2…
Jungkyu Park retweeted
Got experience in #biostatistics & interest in working w/ an interdisciplinary team on leading-edge medical imaging science?
We'd like to hear from you.
#job: cai2r.net/jobs/radiology-bio…
Know someone qualified who may be interested? RT!
#hiring #stats #research #medicine #radiology
Jungkyu Park retweeted
Does your multimodal model fail to utilize all modalities? It could be due to incidental correlation, which is a spurious correlation induced by a lack of data. Learn more in our @TmlrOrg paper w/ @yixinwang_, @kjgeras, and @kchonyc.
openreview.net/forum?id=QoRo…
1/7
Jungkyu Park retweeted
I'm excited to share that the results of the Digital Breast Tomosynthesis Tumor Detection Challenge (DBTex) have officially been published in our @JAMANetworkOpen paper, with our public benchmark, dataset, and detection algorithms' code (links 👇)! jamanetwork.com/journals/jam… (1/n)
Jungkyu Park retweeted
I am happy to share our paper, “Exploring the Acceleration Limits of Deep Learning Variational Network–based Two-dimensional Brain MRI,” which has been recently published in Radiology: Artificial Intelligence.
pubs.rsna.org/doi/10.1148/ry…
Jungkyu Park retweeted
It's typical to sum the task losses in multitask learning (MTL), but this induces spurious dependencies between the input and targets. In our #NeurIPS2022 paper, we mitigate this at inference time using existing MTL methods.
w/ @kchonyc @kjgeras
nips.cc/Conferences/2022/Sch…
1/8
Jungkyu Park retweeted
Introducing HyperTuning: Using a hypermodel to generate parameters for frozen downstream models.
This allows us to adapt models to new tasks *without* back-prop!
Paper: arxiv.org/abs/2211.12485
1/10
Jungkyu Park retweeted
A #DeepLearning algorithm that boosts the accuracy of dynamic contrast-enhanced #MRI for #BreastCancer could reduce false positives and minimize unnecessary biopsies in the clinic, suggests a new study. @JanWitowski @kjgeras @nyugrossman scim.ag/3d
Jungkyu Park retweeted
Our work on AI for breast MRI is finally out in Science Translational Medicine (and on its cover as well👀)
Jungkyu Park retweeted
Check out our #ICML2022 paper!
proceedings.mlr.press/v162/w…
It presents a new perspective on why neural networks struggle to learn effectively from multiple input modalities and provides a simple remedy.
This is a collaboration between @NanWu__, @kudkudakpl, @kchonyc and myself.
Jungkyu Park retweeted
Awesome work from @JeanRemiKing and team shows that activities in the layer hierarchy of a transformer trained with Self-Supervised Learning on speech and audio correlates well with activities of the hierarchy of areas on the human auditory cortex.
🔥Preprint out:
`Toward a realistic model of speech processing in the brain with self-supervised learning’:
arxiv.org/abs/2206.01685
by J. Millet*, @c_caucheteux* and our wonderful team:
The 3 main results summarized below 👇
Jungkyu Park retweeted
Do you know that multi-modal DNNs trained with vanilla SGD favor one of the input modalities and yield limited generalization? Check our paper at #ICML2022 where we characterize this phenomenon quantitatively and provide methods to overcome it: arxiv.org/pdf/2202.05306
Jungkyu Park retweeted
Now published at @SciReports! nature.com/articles/s41598-0…
Thank you again to all of my coauthors across @NYUDataScience, @CILVRatNYU, @NYUImaging, @cai2r, and @PW_edu.
DNNs perform well on a range of medical diagnosis tasks, but do they diagnose similarly to humans?
In breast cancer screening, DNNs use different features than radiologists. Some are spurious, while others may represent new biomarkers.
arxiv.org/abs/2011.14036
1/9
Jungkyu Park retweeted
Happy to share our recent paper “Adaptive Early-Learning Correction for Segmentation from Noisy Annotations” which accepted as Oral presentation at #CVPR2022. Check the paper and code for more information:
abs: arxiv.org/abs/2110.03740
