For this week’s #thicktrunktuesday we’re off to #lebanon and this 600 year old olive tree.

#SnootBoopTuesday #guineapigs

Bishop's Beach in Homer, #Alaska yesterday morning.
📸 by Edward Marsh

"Continual Transformers: Redundancy-Free Attention for Online Inference. (arXiv:2201.06268v3 [cs.AI] UPDATED)" — A novel formulations of the Scaled Dot-Product Attention, which enable Transformers to perform efficient online token-by-token inference on a continual input stream.

Paper: http://arxiv.org/abs/2201.06268
Code: https://github.com/lukashedegaard/continual-transformers

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Multi-block Continual Transform…

"Model soups to increase inference without increasing compute time. (arXiv:2301.10092v1 [cs.CV])" — A comparison of Model Soups performances on three different models (ResNet, ViT and EfficientNet) using three Soup Recipes (Greedy Soup Sorted, Greedy Soup Random and Uniform soup) and a new Soup Recipe called Pruned Soup with better performance.

Paper: http://arxiv.org/abs/2301.10092
Code: https://github.com/milo-sobral/modelsoup

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
A table showing the accuracies …

"Side Eye: Characterizing the Limits of POV Acoustic Eavesdropping from Smartphone Cameras with Rolling Shutters and Movable Lenses. (arXiv:2301.10056v1 [cs.CR])" — A paper discussing the limits of acoustic information leakage caused by structure-borne sound that perturbs the POV of smartphone cameras.

Paper: http://arxiv.org/abs/2301.10056

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Illustration of the POV optical…

"Progressive Meta-Pooling Learning for Lightweight Image Classification Model. (arXiv:2301.10038v1 [cs.CV])" — A framework to make the receptive field learnable for a lightweight network, unlike the conventional efficient learning methods which ignore the role of the receptive field in neural network design, by using parameterized pooling-based operations.

Paper: http://arxiv.org/abs/2301.10038

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Overview of the proposed Meta-P…

@AngelaPreston Now that I can get behind 💯 😃

@AngelaPreston Yep, it looks as if they are trying to detect objects which emit a smell, but using “olfactory” to describe that seems just …. wrong somehow 🙂

"Transfer Learning for Olfactory Object Detection. (arXiv:2301.09906v1 [cs.CV])" — An investigation of the effect of style and category similarity in multiple datasets used for object detection pretraining. The paper discusses using object recognition as a basis for the indirect recognition of more complex and meaningful "smell" references such as gestures and spaces.

Note: Similar to another paper, the word "olfactory" in the title confused me. Is "olfactory" being used in a slightly different context in machine learning?

Paper: http://arxiv.org/abs/2301.09906

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Smell. The Five Senses. 1558 – …

"ODOR: The ICPR2022 ODeuropa Challenge on Olfactory Object Recognition. (arXiv:2301.09878v1 [cs.CV])" — The challenge aims to foster the development of object detection in the visual arts and to promote an olfactory perspective on digital heritage. It appears that they concentrate on smell-relevant objects rather than using smell to detect objects ...

Note: The use of the word "olfactory" is a bit confusing, in my opinion, in this paper and one other which came out today.

Paper: http://arxiv.org/abs/2301.09878

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Example image from the challeng…

A total of 60 papers in the cs.CV category on arXiv.org today — 37 new, 23 updated.

Time for the ol' reading spree ...

#AI #CV #NewPapers #DeepLearning #MachineLearning

#CogSci #CognitiveScience #psychology #OpenScience

In the 2023/2024 academic year, we're launching a new Master's in Cognitive Science at the University of Rijeka. It's an interdisciplinary research-intensive course in English, pedagogically innovative, comparatively cheap (both for EU and non-EU students), accredited, and in beautiful Rijeka!

https://cogsci.uniri.hr/

@DaNanner Well, Laurie’s currently QA but I’m sure we can expand the QA department 😛

I like that Tusker supports Akkoma out of the box (I had to add that to IceCubes but I’d been working with IceCubes code for a while to get it working on Akkoma and so that wasn’t a huge issue) but Tusker lacks edit functionality while IceCubes has it now …

Tusker has lists and hashtags support but doesn’t seem to pull any hashtags you follow now from anywhere (I haven’t looked at Akkoma API to see if this is possible though ..) while IceCubes has no support for those or even for filtering.

Tusker also has this nifty feature which combines all the boosts for a post in one notification that I haven’t seen anywhere else, but their CW feature seems a bit weird 😛

Source code for either is easy to work with and so I’ve punted the decision as to which one to go with to Laurie and am waiting …

Tusker is an excellent Mastodon iOS app that has just moved from beta onto the Apple app store. Check it out. It’s free. And it’s very good. #mastodon #Tusker

@DaNanner Sorry to hear about being overwrought. Hope things get better for you on that front …

Regarding a Mastodon client, Tusker and IceCubes both look great and have different features that I’d want. Plus, they’re both open source (as is MetaText if I’m not mistaken) …

Laurie is looking at Tusker and IceCubes to see which one she likes and then I’m hoping to start adding to the one she likes to create the perfect Mastodon client for our own needs since the code is available. Would be happy to share whatever we end up with if you’d like to try …

New call for pitches from MIT Technology Review, for our upcoming issue on #accessibility. We're looking for *big* swings: narrative features, investigations, essential profiles, and sharp essays at the general intersection of accessibility and technology.

Pitch deadline: Feb 3
Our rates range from $1-2/word.

This is the first time I've posted a pitch call to @Mastodon, so let me know if there are hashtags I should add or groups I should cc. #freelance #journalism #technology #a11y

It's raining on and off; little chance to indulge in gardening today.

Got a few photos and now back indoors just as rain picked up. It's more than a drizzle now - the plants seem delighted to be spared from cycles of frost and baking sun.

#gardening #Kodaikanal

The dock at Douglas Island with a backdrop of snowy mountains in Juneau, #Alaska
📸 by Viliame Kaulotu

#artspam #clouds #sky