Bio-PrecisionAI Health

01/28/2026

01/26/2026

Layers of AI

#1. Classical AI (Rule-Based AI)
What it is: The earliest form of AI, based on explicit rules written by humans.

components:
- Symbolic AI
- Expert Systems
- Knowledge Representation
- Logic & Reasoning
No learning from data & Rigid and hard to scale

# 2. Machine Learning (ML)
What it is: Systems learn patterns from data instead of fixed rules.

Key components:
Supervised Learning | Unsupervised Learning | Reinforcement Learning
Classification & Regression

#3. Neural Networks
What it is: ML models inspired by the human brain, using layers of neurons.

# 4. Deep Learning
What it is: Neural networks with many layers, capable of high-level feature learning.

Key architectures:
CNNs → images & vision
RNNs / LSTMs → sequences & time series
Transformers → language & multimodal data
Autoencoders → representation learning

Example:
Face recognition, speech-to-text, translation.

#5. Generative AI
What it is: Models that create new content, not just predict.

Key models:
LLMs (text generation)
Diffusion Models (image/video generation)
VAEs
Multimodal Models (text + image + audio)
Example: Chatbots, image generators, code assistants.
Key leap: AI becomes creative, not just analytical.

#6. Agentic AI (Autonomous AI)
What it is: AI systems that plan, remember, decide, use tools, and act autonomously.
Core capabilities:
Memory
Planning
Tool Use (APIs, databases, code ex*****on)
Autonomous Ex*****on

By KODI PRAKASH SENAPATI

01/24/2026

Scaling Text-to-Image Diffusion Transformers with Representation Autoencoders

Shengbang Tong, Boyang Zheng, Ziteng Wang, Bingda Tang, Nanye Ma, Ellis Brown, Jihan Yang, Rob Fergus, Yann LeCun, Saining Xie

New York University 2026
https://arxiv.org/abs/2601.16208

Using a simpler design to unlock better AI image generation.

By operating in a shared, high-dimensional semantic representation space, Representation Autoencoders (RAEs) allow diffusion architectures to be simplified at scale.

In controlled comparisons against the FLUX VAE, RAE-based diffusion transformers consistently converge faster, avoid catastrophic overfitting during finetuning, and achieve higher-quality text-to-image generation across model sizes from 0.5B to 9.8B parameters.

Notes:

Modern text-to-image systems—those that turn written prompts into pictures—usually work by compressing images into a hidden “latent” space, generating new content there, and then decoding it back into pixels. The quality of that hidden space matters enormously. This paper asks a simple but consequential question: are we using the right kind of latent representation as these models scale up?

Traditionally, most large text-to-image models rely on *variational autoencoders* (VAEs) to create this latent space. But recent work on image classification hinted that a different approach—*representation autoencoders* (RAEs)—might offer cleaner, more semantic representations. The authors of this study explore whether RAEs can handle the far messier world of large-scale, freeform text-to-image generation.

To test this, they scale RAEs well beyond curated datasets like ImageNet, training them instead on a mix of web images, synthetic data, and images containing text. They find that simply making models bigger improves overall image quality—but that *what data you train on still matters*, especially for tricky domains like rendering readable text in images.

The team then takes a hard look at the design tricks that previously made RAEs work on smaller datasets. Surprisingly, many of these complexities turn out to be unnecessary at scale. As models grow, the system actually becomes simpler: only the way noise is scheduled during diffusion remains critical, while other architectural embellishments add little value.

With this streamlined setup, the researchers run a head-to-head comparison between RAEs and a leading VAE-based system (FLUX), scaling models from hundreds of millions to nearly ten billion parameters. The results are striking. Across all sizes, RAE-based models learn faster, produce better images during pretraining, and—crucially—remain stable during long finetuning runs. In contrast, VAE-based models begin to overfit and collapse after extended training, even on high-quality datasets.

Beyond better images, RAEs offer a deeper advantage. Because both image understanding and image generation happen in the *same representation space*, the model can directly reason about what it generates—opening the door to systems that don’t just create images, but can also think about them in a unified way.

Taken together, the results suggest a shift in foundation: representation autoencoders are not just a viable alternative to VAEs—they may be a simpler, more robust, and more scalable backbone for the next generation of text-to-image models.

Representation Autoencoders (RAEs) have shown distinct advantages in diffusion modeling on ImageNet by training in high-dimensional semantic latent spaces. In this work, we investigate whether this framework can scale to large-scale, freeform text-to-image (T2I) generation. We first scale RAE decode...

01/23/2026

These hidden Al tools make studying 10x faster.

1. Perplexity.ai - Find answers

2. Elicit.org - Find research papers

3. Scispace.com - Read papers

4. Explainpaper.com - Explain hard papers

5. Humata.ai - Ask PDFs

[Save]

Copied

Humata is an AI agent that turns your documents into a fast, intelligent knowledge base for instant analysis, insights, and answers.

01/19/2026

“We must remember that intelligence is not enough. Intelligence plus character — that is the goal of true education.” ~ Martin Luther King, Jr.

01/17/2026

01/16/2026

From Entropy to Epiplexity: Rethinking Information for
Computationally Bounded Intelligence

Instead of entropy, they introduced a new measure—known as Epiplexity. It can potentially show if a data will help your model.

Read more here: https://arxiv.org/pdf/2601.03220

….

01/08/2026

ConvBERT

Overview

The ConvBERT model was proposed in ConvBERT: Improving BERT with Span-based Dynamic Convolution by Zihang Jiang, Weihao Yu, Daquan Zhou, Yunpeng Chen, Jiashi Feng, Shuicheng Yan.

The abstract from the paper is the following:

Pre-trained language models like BERT and its variants have recently achieved impressive performance in various natural language understanding tasks. However, BERT heavily relies on the global self-attention block and thus suffers large memory footprint and computation cost. Although all its attention heads query on the whole input sequence for generating the attention map from a global perspective, we observe some heads only need to learn local dependencies, which means the existence of computation redundancy. We therefore propose a novel span-based dynamic convolution to replace these self-attention heads to directly model local dependencies. The novel convolution heads, together with the rest self-attention heads, form a new mixed attention block that is more efficient at both global and local context learning. We equip BERT with this mixed attention design and build a ConvBERT model. Experiments have shown that ConvBERT significantly outperforms BERT and its variants in various downstream tasks, with lower training cost and fewer model parameters. Remarkably, ConvBERTbase model achieves 86.4 GLUE score, 0.7 higher than ELECTRAbase, while using less than 1/4 training cost. Code and pre-trained models will be released.

This model was contributed by abhishek. The original implementation can be found here: https://github.com/yitu-opensource/ConvBert

Usage tips

ConvBERT training tips are similar to those of BERT. For usage tips refer to BERT documentation.

Copied

Contribute to yitu-opensource/ConvBert development by creating an account on GitHub.

01/07/2026

How does Altitude Affect AS Genotype?

The term altitude, in relation to an AS genotype, refers to how the environmental stress of high elevation (specifically the low oxygen levels, or hypoxia) interacts with individuals who carry the gene for sickle cell trait (AS genotype).

The key points regarding altitude and the AS genotype are:

Hypoxia Challenge: High altitude environments (typically defined as above 2,500 meters or ~8,200 feet) present a major physiological challenge due to reduced ambient oxygen. The human body responds to this by acclimatizing, often by producing more red blood cells (erythrocytosis).

Sickle Cell Trait (AS Genotype): People with the AS genotype are heterozygous for the sickle cell allele, meaning they have one copy of the normal hemoglobin gene (A) and one copy of the sickle cell gene (S). Generally, these individuals are healthy and do not experience severe symptoms at sea level.

Maladaptation at Altitude: The unique physiological changes that occur at high altitude, particularly the lower oxygen saturation and changes in blood flow/viscosity, can be problematic for individuals with the AS trait. The lower oxygen tension can trigger the red blood cells to sickle more readily, even in those who are only carriers.

Increased Risk: For people with the AS genotype, exposure to high altitudes can increase the risk of altitude-related illnesses and other health complications, such as:
Vaso-occlusive events: Sickled red blood cells can block small blood vessels, leading to pain and organ damage.

Splenic syndrome: The spleen may become enlarged and painful.
Deep venous thrombosis (DVT): There is an increased risk of blood clots.

Therefore, from a medical and biological perspective, altitude represents an environmental stressor that can turn an otherwise benign AS genotype into a health risk, unlike the high-altitude adaptations seen in indigenous populations (like Tibetans and Andeans) who have evolved distinct genetic mechanisms for coping with chronic hypoxia.

Copied

01/07/2026

What Is Sickle Cell Trait (AS genotype)?

Sickle cell trait is not a disease. Sickle cell trait is the inheritance of one gene for sickle hemoglobin and one for normal hemoglobin. Sickle cell trait will not turn into the disease. Sickle cell trait is a life-long condition that will not change over time.
• During intense exercise, red blood cells containing the sickle hemoglobin can change shape from round to quarter-moon, or “sickle.”
• Sickled red cells may accumulate in the blood stream during intense exercise, blocking normal blood flow to the tissues and muscles.
• During intense exercise, athletes with sickle cell trait have experienced significant physical distress, collapsed and even died.
• Heat, dehydration, altitude and asthma can increase the risk for and worsen complications associated with sickle cell trait, even when exercise is not intense.
• Athletes with sickle cell trait should not be excluded from participation as precautions can be put into place.

How Can I Prevent a Collapse?

• Know your sickle cell trait status.
• Engage in a slow and gradual preseason conditioning regimen.
• Build up your intensity slowly while training.
• Set your own pace. Use adequate rest and recovery between repetitions, especially during “gassers” and intense station or “mat” drills.
• Avoid pushing with all-out exertion longer than two to three minutes without a rest interval or a breather.
• If you experience symptoms such as muscle pain, abnormal weakness, undue fatigue or breathlessness, stop the activity immediately and notify your
athletic trainer and/or coach.
• Stay well hydrated at all times, especially in hot and humid conditions.
• Avoid using high-caffeine energy drinks or supplements, or other stimulants, as they may contribute to dehydration.
• Maintain proper asthma management.
• Refrain from extreme exercise during acute illness, if feeling ill, or while experiencing a fever.
• Beware when adjusting to a change in altitude, e.g., a rise in altitude of as little as 2,000 feet. Modify your training and request that supplemental oxygen be available to you.
• Seek prompt medical care when experiencing unusual physical distress.

For more information and resources, visit ncaa.org/health-safety.

Sickle cell
Healthy red
blood cell

Do You Know If You Have Sickle Cell Trait?

People at high risk for having sickle cell trait are those whose ancestors come from Africa, South or Central America, India, Saudi Arabia and Caribbean and Mediterranean countries.
• Sickle cell trait occurs in about
8% of the U.S. African American
population, and between one in
2,000 and one in 10,000 in the
Caucasian population.
• All NCAA student-athletes
must provide their school with
documented results from a
previous sickle cell solubility test,
or they must undergo testing
during their preparticipation
medical examination.
• Knowledge of sickle cell trait status can be a gateway to education and simple precautions that may prevent collapse among athletes with sickle cell trait, allowing you to thrive in your sport.
• All newborns in the United States
(and newborns in many other
countries) undergo sickle cell
testing at birth.

Copied

The official Health, Safety & Performance page for the NCAA.org