Well done, @zhangdkai
presenting your research "Targeted Activation Penalties Help CNNs Ignore Spurious Signals” at #AAAI24. #AI4HealthCentre #UKRI

Our #AAAI24 poster today is about turning gold into blue and changing a 3 into a 5, all with the help of the energy that trained models happen to collect anyway. ECCCos FTW! (Poster nr 514)

@patalt has a great blog post explaining it all, with amazing pictures that are also on the poster: https://www.paltmeyer.com/blog/posts/eccco/

Heading to #AAAI24 today with for the XAI for DRL workshop: xai4drl.github.io. Let me know if you're around!

V-JEPA: The next step toward Yann LeCun’s vision of advanced machine intelligence (AMI) — https://ai.meta.com/blog/v-jepa-yann-lecun-ai-model-video-joint-embedding-predictive-architecture/

#AAAI24

Yann LeCun has some recommendations in his #AAAI24 keynote:

• Abandon generative models in favor joint-embedding architectures
• Abandon probabilistic model in favor of energy-based models
• Abandon contrastive methods in favor of regularized methods
• Abandon Reinforcement Learning in favor of model-predictive control; Use RI only when planning doesn't yield the predicted outcome, to adjust the world model or the critic

Very impressed by Milind Tambe's #AAAI24 keynote on "AI for Social Good".

A packed series of impactful projects, covering fighting HIV in LA, wild life conversation in Africa, mobile health programs for maternal and child in India, and. a lot more.

All projects face scarcity of resources, for which Tambe manages to find (extensions of) restless multi-armed bandits as solution.

https://aaai.org/aaai-conference/invited-speakers/#Tambe

#MilindTambe #AIforSocialGood #AAAI

Ouch: In #AAAI24 session where in two consecutive oral paper presentations the speakers didn't show up 😟 .

All authors really want to present their paper, but only 10% of the accepted papers are allowed to. And then they don't come?

At the opening event of #AAAI24, the 38th Annual AAAI Conference on Artificial Intelligence, in Vancouver.

The scale of this AI conference is quite different from what I'm used to in software engineering. 2,400 accepted papers (out of 10,000 submissions), all presented as posters in three nights with 800 posters each, and a subset of 240 papers presented in "orals". And 5,000 participants.

Getting ready for being completely overwhelmed.

https://aaai.org/aaai-conference/

The amount information I’m exposed to at academic conferences always feels slightly overwhelming. I have changed and then overthrown plans for my next thesis chapter three times today. Help #aaai24

📢 Join us for the #AAAI24 Spring Symposium on Human-Like Learning, March 25-27th at Stanford!

Many #AI / #ML systems can achieve human-like performance, but come no where close to exhibiting human-like learning. At the symposium, we’ll discuss how to build AI systems that exhibit human-like learning capabilities, such as online, incremental, continual, interactive, and data/compute efficient learning.

👉 Interested? Submit a talk abstract by January 10th. See: https://humanlikelearning.com/aaai24-ss/

Good news! We have one paper accepted at #aaai24

The paper is titled "Neural Reasoning About Agents’ Goals, Preferences, and Actions" and it introduces the Intuitive Reasoning Network (IRENE) – a new neural model for intuitive psychological reasoning.

Authors: Matteo Bortoletto, Lei Shi, @abulling

The preprint and the code are available at https://www.perceptualui.org/publications/bortoletto24_aaai/.

Good news: our paper “Faithful Model Explanations through Energy-Constrained Conformal Counterfactuals” (which we abbreviate as ECCCos) was accepted at #AAAI24.

Preprint still in the making but you can get a sneak preview in this presentation on “ECCCos from the Black Box” at the launch of the Delft Fintech Lab in May this year. https://www.paltmeyer.com/content/talks/posts/2023-delft-fintech/

With @patalt, @cynthiacsliem, Mojtaba Framanbar