Lo vi y lo estuve estudiando meter en el vídeo PEEEEEERO tiene un poco de trampa. Comparar modelos unimodales de texto (ciegos como GPT 3.5) en tareas de QyA visuales es un poco raro. Y más teniendo la opción de validarlo contra BLIP.
RESEARCH
-
MIT researchers deploy AI for earthquake relief in Turkey Syria
By
–
New MIT Tech Review article on how @Ritwik_G and collaborators @berkeley_ai @DIU_x @Microsoft are currently deploying AI for earthquake relief efforts in Turkey and Syria.
-
Future Works Should Document AI Drawbacks and Limitations
By
–
It sounds like the responsibility of future works to mention these drawbacks, so that hopefully it becomes well known.
-
Combinatorial Optimization and RL Shape Future Game Design
By
–
Overall, cool paper about a funny application of combinatorial optimization. Jokes aside, RL and optimization are probably the future of game design. At least for puzzle games. 🙂
-
Genetic Algorithm for Puzzle Solving with Diversity Selection
By
–
Next, solving puzzles. The authors used a custom GA: 1/ Initialization using the previous algorithm 2/ Crossover/Mutation with a uniform crossover operation 3/ Find feasible solutions and "heal" broken ones 4/ Selection with a diversity criterion
-
Multi-Island Genetic Algorithm for Diversity Maintenance
By
–
Maintaining diversity is a big focus, so they implemented a "multi-island" GA. → Each population is divided into several sub-populations called "islands" → Applies GA separately on each sub-population, periodically migrating individuals between them
-
Multi-Island Genetic Algorithms Outperform Standard Approaches
By
–
Here are the final results. The multi-island outperforms the standard GA while maintaining higher diversity. There are a lot of useful tips and tricks about GAs in this paper. GAs deserve more love.
-
Graph Node Selection Algorithm With Constraint-Based Synergy Optimization
By
–
Proposed solution: 1/ Defining a random traversal order for graph nodes 2/ Filtering the list of candidates according to constraint requirements 3/ Selecting an item that maximizes the synergy of partially filled neighboring candidates.
-
Building High-Synergy Agent Formations and Optimal Puzzle Solutions
By
–
The authors divide this problem into 2 subproblems: 1/ Building formations with high synergy 2/ Solve puzzles in a near-optimal way
-
Optimizing Puzzle Generation: Beyond Exhaustive Search Methods
By
–
Current landscape: • Traditional approaches to generating puzzles are often time-consuming • Exhaustive search may not be feasible due to the large design space