Brian Jalaian

🚀 Associate Professor @UWF | Researching LLMs, Agentic AI, and AI Optimization at the Edge | Building the future of robust, efficient AI.

[Brian Jalaian @BJalaian]

What if you could talk to your drones?

We’re building an AI-enabled autonomy stack with WebTAK, edge AI, open-source VLMs, and visual reasoning.

Full demo 👇
https://t.co/FeJrvb3CN9
#PhysicalAI #EdgeAI #Robotics

[Brian Jalaian @BJalaian]

@WhiteHouse Thank you President Trump

[Brian Jalaian @BJalaian]

@danvahdat @elonmusk Thank you @elonmusk

[Brian Jalaian @BJalaian]

🚀 Want to train ML models with free GPUs in the cloud?

In just 2 mins, David Moe from Jalaian AI Lab shows how we teach:

✅ Google Colab + Kaggle
✅ AI-assisted coding (Gemini)
✅ Easy submission.csv for competitions

🎥 Watch full video: https://t.co/7OLfJtyXmL
#AICoding

[Brian Jalaian @BJalaian]

“Don’t fear uncertainty — that’s where the magic happens.”
Shared a 3-min clip from a convo with Arman on books, mindset & the startup journey.
▶️ Watch:https://t.co/RrwIgAS7yj
#entrepreneurship #growthmindset #startuplife #booktalk

[Brian Jalaian @BJalaian]

Year 3 of our Machine Learning in Data Science course is here—with a new focus on AI tools like Copilot 🤖⚡ Excited to explore how these tools can boost learning and creativity. Big shoutout to the research team pushing boundaries! #MachineLearning #AI #DataScience

[Brian Jalaian @BJalaian]

We trained ML models to detect cognitive load using wearable biosignal data.
👨‍⚕️ Wristbands + EEG → Features + Spectrograms → Transfer Learning
Here’s a 3-min clip breaking it all down:
▶️ https://t.co/4LG8dcNwic
#DeepLearning #Neurotech #Biosignals #TechForGood

[Brian Jalaian @BJalaian]

🧠 Can AI detect your mental effort using wearables?

We tested ML, DNNs & transfer learning on biosignal data (EEG, PPG, BVP) to decode cognitive states.
📺 Full video: https://t.co/yqPCIP1kiy

#AI #DeepLearning #Wearables #CognitiveLoad #MachineLearning

[Brian Jalaian @BJalaian]

🎥 New 3-min explainer:
Visual Reasoning Agent (VRA) = training-free agentic AI for vision systems 💡

Up to +40% accuracy on tough remote sensing tasks — no retraining, just smart reasoning.
Think. Critique. Act.

📽️ Watch how VRA boosts robustness👇
#VLM #AgenticAI

[Brian Jalaian @BJalaian]

@sama @Sama 🔥 Practicing o3 3+ hrs/day—what learning path works best? Structured tutorials vs OSS deep dives? 📚 vs 🛠️ balance? Tips on feedback loops or AI/human mentors? How do YOU use AI to level up your #AIresearch game? #skillsmax #o3

[Brian Jalaian @BJalaian]

🌟 Hydra shows that structured, agentic reasoning can dramatically improve VLM reliability efficiently.
We’re excited about future work combining Hydra with even larger LLMs and edge deployment!
📄 https://t.co/nNfT37rafZ
#AI #LLMs #RobustAI #VisionLanguageModels

[Brian Jalaian @BJalaian]

🚀 Excited to share our new research! We present Hydra, boosting adversarial robustness and reducing hallucinations in VLMs.

Hydra enhances reliability through iterative reasoning—no extra training needed.

📄 https://t.co/nNfT37rafZ

#AI #VLM #Robustness #ML #Research

[Brian Jalaian @BJalaian]

🧪 Results:
Hydra outperforms existing plug-in methods across benchmarks like ScienceQA, GQA, and TextVQA.
It’s particularly effective against unseen adversarial examples — even without explicit defenses.

[Brian Jalaian @BJalaian]

🏋️‍♂️ Why is this important?
Stronger resilience against adversarial attacks
Reduced hallucination rates
Better factual accuracy in open-world settings
Works without expensive retraining or fine-tuning!

[Brian Jalaian @BJalaian]

It uses Chain-of-Thought (CoT) and In-Context Learning (ICL) techniques to generate, critique, and refine answers across multiple reasoning passes.
Hydra acts like an "agent," adapting based on feedback.

[Brian Jalaian @BJalaian]

Vision-Language Models are incredibly powerful but vulnerable to adversarial attacks and hallucinations.
Hydra tackles both issues by introducing an iterative reasoning framework — no retraining needed.

[Brian Jalaian @BJalaian]

🚨 Join the AAAI FS 2024 - ATRACC for expert talks on AI Validation, Risk, & Trustworthiness! 🗓️ Nov 7-9, 2024 📍 Arlington, VA 🔑 Hotel Deadline: Oct 17 – Limited rooms! 🏨 More info: https://t.co/4OzFqp24ga #AAAI #ATRACC2024 #AITrustworthiness #AIConference 🚨

[Yann LeCun @ylecun]

Yes, many of us have always been working on "AGI", whatever you mean by that.
Perceptron, General Problem Solver, expert systems, machine learning, backprop, RL, SSL, transformers, LLMs...: For some, these were going to be "The One Weird Trick" that was gonna take us to AGI.
1/

[Ethan Mollick @emollick]

Extraordinary new paper from Google on medicine & AI: When Google tuned a AI chatbot to answer common medical questions, doctors judged 92.6% of its answers right … compared to 92.9% of answers given by other doctors.

And look at the pace of improvement! https://t.co/2Tc2YTUR6E

[Yann LeCun @ylecun]

Neural nets that analyze/generate text, speech, image, proteins don't tell us much about the structure thereof.
But
1. they are very useful
2. they tell us that SGD is better at structure discovery than humans
3. they upset people who devoted their career to structure discovery