Posts
1411
Following
142
Followers
869
I'm a bit of an eclectic mess 🙂 I've been a programmer, journalist, editor, TV producer, and a few other things.

I'm currently working on my second novel which is complete, but is in the edit stage. I wrote my first novel over 20 years ago but then didn't write much till now.

I post about #Coding, #Flutter, #Writing, #Movies and #TV. I'll also talk about #Technology, #Gadgets, #MachineLearning, #DeepLearning and a few other things as the fancy strikes ...

Lived in: 🇱🇰🇸🇦🇺🇸🇳🇿🇸🇬🇲🇾🇦🇪🇫🇷🇪🇸🇵🇹🇶🇦🇨🇦

Fahim Farook

"TetCNN: Convolutional Neural Networks on Tetrahedral Meshes. (arXiv:2302.03830v1 [cs.CV])" — A novel interpretable graph Convolutional Neural Network (CNN) framework for tetrahedral mesh structures, inspired by ChebyNet.

Paper: http://arxiv.org/abs/2302.03830

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
TetCNN architecture for the cla…
0
2
1

Fahim Farook

"How to Trust Your Diffusion Model: A Convex Optimization Approach to Conformal Risk Control. (arXiv:2302.03791v1 [stat.ML])" — A focus on image-to-image regression tasks to arrive at a generalization of the Risk-Controlling Prediction Sets (RCPS) procedure, which allows to provide: i) entrywise calibrated intervals for future samples of any diffusion model, ii) control a certain notion of risk with respect to a ground truth image with minimal mean interval length.

Paper: http://arxiv.org/abs/2302.03791

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Example images
0
0
0

Fahim Farook

"Understanding Why ViT Trains Badly on Small Datasets: An Intuitive Perspective. (arXiv:2302.03751v1 [cs.CV])" — A visual intuition to help understand why ViT has a significantly lower evaluation accuracy when trained on small datasets when compared to ResNet-18 with a similar number of parameters.

Paper: http://arxiv.org/abs/2302.03751
Code: https://github.com/BoyuanJackChen/Visualize-Transformer-ResNet18

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Visualization for ViT on CIFAR-…
0
1
0

Fahim Farook

"Towards causally linking architectural parametrizations to algorithmic bias in neural networks. (arXiv:2302.03750v1 [cs.CV])" — A causal framework for linking an architectural hyperparameter to algorithmic bias to study effect of hyperparameters on inducing biases.

Paper: http://arxiv.org/abs/2302.03750

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
The deep learning model develop…
0
1
0

Fahim Farook

Looks as if the #StableDiffusion image generation on the #Swift side of the fence for #macOS has been progressing faster than I thought 🙂

It feels as if it’s been barely any time since I last took a look but there are a lot more #CoreML StableDiffusion models out there now and a lot more GUIs too. I took a look at a few, combined features from a few, added my own features too and came up with something which lets me maintain my image gallery and to generate new images too.

Given that on my 32GB M1 MBP it takes about 13 seconds to generate an image at 20 steps (except for the first time) it is pretty fast too. At least, fast enough for my needs 🙂

So it looks as if I’m back to messing with StableDiffusion on Swift rather than #Python, at least for the time being …
Showing a GUI for StableDiffusi…
1
0
5

Fahim Farook

Edited 1 year ago
Long text - serialized story
Show content
Here’s part #1 in a story I hope to serialise weekly on here:

The Basement of the Universe — Part 1

“Welcome to the Basement of the Universe!”

DoorBar forced his eyes open with the kind of effort you would normally put into opening the heavy, creaking gates of an old castle. What was going on? Where was he?

All the other questions that were jostling for attention in his head disappeared as he saw who’d just spoken to him. He was now sure that he was still asleep since the creature before him was unlike anything he’d seen in his waking moments.

It looked like a rabbit going full tilt had crashed into an egg --- a huge egg at that. It had rabbit ears, a face that looked kind of like a rabbit’s, but a body that was definitely --- or maybe defiantly? --- egg shaped.

What kind of weird nightmare was this?

DoorBar decided that he might as well make the best of it, if it was indeed a dream or nightmare, and find out what was going on. At least, it would make a good story to tell people ... “If I remember when I wake up,” he reminded himself.

“Who are you?” he asked, looking at the creature.

“Oh, I’m sorry. Where are my manners? I’m a hopper, my name is Kluk.”

That started a whole bunch of other questions jousting for dominance in the theater of DoorBar’s mind. He put his mind on hold, and let his mouth take over.

“You’re not a hopper! And hoppers don’t talk!”

The creature, what was his name, Cluck?, tipped it’s head to one side and looked at DoorBar with wide eyes. Could it’s eyes get any wider, he idly wondered.

“Why am I not a hopper?”

“A hopper is round, white, and sometimes has an egg in the center.”

Kluk gestured at his --- or, her? His mind prompted --- body as if it was evidence enough and said, “And?”

“No, no,” said DoorBar trying to figure out how best to explain things to the creature.

“You eat hoppers!” he babbled.

Hopper still looked at him as if this was the most ordinary thing in the world. “Maybe in this world it is?” he speculated.

“Sure, some people do eat hoppers. Personally, I think it’s a little barbaric, but who am I to argue about other people’s dietary habits?”

“No, I mean ... well ... you ... I ... Where I come from, hoppers are an item of food. They make it using umm ... flour or batter maybe? It’s small, you can hold it in your hand ... “ DoorBar flailed, trying to dredge up any vestiges of culinary knowledge to describe hoppers --- but alas, he was a consumer, not a chef and he’d reached the limits of his hopper knowledge.

“Ah,” nodded Kluk sagely. “They do have lots of weird things in the universe and we get all sorts of people here. So I guess it makes sense that there’d be other kinds of hoppers ...”

DoorBar struggled to pay attention to what Kluk was saying, but his mind would have none of it. It was too intent on another question and again, his mouth took over.

“What sort of name is Cluck? Sounds like a chicken!” he blurted out.

“It’s not Cluck!” replied the creature, with a tiny hint of heat. “It’s Kluk. There’s a difference in the sounds,” it said, as if talking to a child. “Most people can recognize the difference between the two. Imagine, if everybody went around mistaking the two ...” it continued, almost as if talking to itself.

DoorBar dismissed Cluck’s --- Kluk’s, he corrected himself --- ramblings and moved on to the next loudest question in the chorus line.

“Where am I?”

“You’re in the Basement.”

“What’s a Basement?”

“Well, you see, sometimes a building will have an underground room for storage and stuff ...

Was the creature messing with him? DoorBar looked at Kluk with narrow eyed. Not that he could really tell what was going on based on the expressions on a rabbit face, but it seemed as if Kluk was responding to his question earnestly. Perhaps he needed to take a different tack?

“No, no, I mean what’s this place? Why do you keep calling it a basement when we are outside?” DoorBar gestured at their surroundings, which, now that he had a better look, stretched on for what seemed like forever. It was all trees, bushes, flowers, butterflies, and blue skies --- almost as if out of a painting. “So, not a nightmare, but a dream then?” he mumbled to himself.

Kluk seemed to have heard his mumblings.

“Dream?” he asked, with a hint of a smile. “Oh no, my friend, this is no dream!” he paused, as if considering. “Sure, a lot of people who arrive in the basement think they are dreaming at first, but this is no dream. It’s all real.”

“There you go again, talking about a basement. What is this basement?”

“Ah, right. Well, you are in the Basement of the Universe” --- DoorBar could almost see the capitalization of the name --- “and you have fallen through a crack in reality into the basement!” said Kluk, almost as if expecting a drumroll, or a banging of a gong, at this point.

Crack? It sounded as if the creature was on crack alright. What did it mean a crack in reality?

“What do you mean a crack in reality?”

“The universe is a big place,” replied Kluk, enunciating clearly and slowly, as if explaining things to a child ... or a very slow adult. “And reality is kind of stretched thin covering everything. So, every once in a while a crack will appear in reality ...”

“A crack?”

“Yes, you know --- a hole, a rip, a tear, whatever you want to call it. Generally, reality will repair itself fairly quickly but there are times when the repair process isn’t fast enough and some items, or people, fall through the crack before the fix is in place.”

“Hmm ...” said DoorBar, part of his mind engaging in all sorts of gymnastics to adapt to the new facts, while the rest of his mind did all it could to deny the information and to prepare its bulwarks against the siege of this new information. “We have had people disappearing everywhere in Colombo recently ...”

“Yes,” said Kluk with the enthusiasm that every teacher shows on seeing the spark of knowledge taking hold and beginning to cast enlightenment, “That’s right! The cracks have been getting more frequent recently!”

But DoorBar had already left the cracks in his wake. The flimsy barricades that his mind had erected against the reality of the situation were already crumbling and he was beginning to realize that this might not be a dream after all. He’d already tried pinching himself --- they said that worked in dreams, but did it? --- and he had not woken up. So did this mean that he was now stuck in this strange Basement, whatever, or wherever, it was?

“And what of Rani?” he asked himself, his thoughts roiling in fresh frenzy as the reality of the situation hardened around him like the concrete vest around a man thrown into the ocean by the mob.

—-

The art is by my wife @Laurie

Annotations (in case you want to find out a bit more info about what is behind some of the elements in the story):

DoorBar - https://en.wikipedia.org/wiki/Darbar_(title)

Kluk - https://en.wikipedia.org/wiki/Kukulkan

Hoppers - https://en.wikipedia.org/wiki/Appam

Colombo - https://en.wikipedia.org/wiki/Colombo

Rani - https://en.wikipedia.org/wiki/Rani

#Writing #AmWriting #Serialized #ScienceFiction
A creature which has an egg-sha…
0
1
1

Fahim Farook

"Name Your Colour For the Task: Artificially Discover Colour Naming via Colour Quantisation Transformer. (arXiv:2212.03434v2 [cs.CV] UPDATED)" — An exploration as to whether machine learning could evolve and discover a colour-naming system similar to humans via optimising the communication efficiency represented by high-level recognition performance.

Paper: http://arxiv.org/abs/2212.03434

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
(a) the theoretical limit of ef…
0
2
0

Fahim Farook

"RegMixup: Mixup as a Regularizer Can Surprisingly Improve Accuracy and Out Distribution Robustness. (arXiv:2206.14502v2 [cs.LG] UPDATED)" — Improving Mixup further by using it as an additional regularizer to the standard cross-entropy loss, instead of using it as the sole learning objective.

Paper: http://arxiv.org/abs/2206.14502
Code: https://github.com/francescopinto/regmixup

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Mixup vs RegMixup in practice. …
0
3
1

Fahim Farook

"Auditing Gender Presentation Differences in Text-to-Image Models. (arXiv:2302.03675v1 [cs.CV])" — A method using fine-grained self-presentation attributes to study how gender is presented differently in text-to-image models to quantify the frequency differences of presentation-centric attributes (e.g., "a shirt" and "a dress").

Paper: http://arxiv.org/abs/2302.03675

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Based on our prompts, we report…
0
0
1

Fahim Farook

"NICER-SLAM: Neural Implicit Scene Encoding for RGB SLAM. (arXiv:2302.03594v1 [cs.CV])" — A dense RGB Simultaneous Localization And Mapping (SLAM) system that simultaneously optimizes for camera poses and a hierarchical neural implicit map representation, which also allows for high-quality novel view synthesis.

Paper: http://arxiv.org/abs/2302.03594

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
3D Dense Reconstruction and Ren…
0
3
1

Fahim Farook

"Sparse Mixture Once-for-all Adversarial Training for Efficient In-Situ Trade-Off Between Accuracy and Robustness of DNNs. (arXiv:2302.03523v1 [cs.CV])" — A method that allows a Deep Neural Network (DNN) to train once and then in-situ trade-off between accuracy and robustness, that too at a reduced compute and parameter overhead.

Paper: http://arxiv.org/abs/2302.03523

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Performance comparison of the p…
0
1
0

Fahim Farook

"ImaginE: An Imagination-Based Automatic Evaluation Metric for Natural Language Generation. (arXiv:2106.05970v2 [cs.CL] UPDATED)" — An imagination-based automatic evaluation metric for natural language generation, which with the help of StableDiffusion, automatically generates an image as the embodied imagination for a text snippet and computes the imagination similarity using contextual embeddings.

Paper: http://arxiv.org/abs/2106.05970

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
An evaluation example on GigaWo…
0
1
0

Fahim Farook

"ShiftDDPMs: Exploring Conditional Diffusion Models by Shifting Diffusion Trajectories. (arXiv:2302.02373v1 [cs.CV])" — A novel and flexible conditional diffusion model by introducing conditions into the forward process. The process utilizes extra latent space to allocate an exclusive diffusion trajectory for each condition based on some shifting rules, which will disperse condition modeling to all timesteps and improve the learning capacity of model.

Paper: http://arxiv.org/abs/2302.02373

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Exploration of the mechanism of…
0
1
0

Fahim Farook

"ReDi: Efficient Learning-Free Diffusion Inference via Trajectory Retrieval. (arXiv:2302.02285v1 [cs.CV])" — A simple yet learning-free Retrieval-based Diffusion sampling framework capable of fast inference, which retrieves a trajectory similar to the partially generated trajectory from a precomputed knowledge base at an early stage of generation, skips a large portion of intermediate steps, and continues sampling from a later step in the retrieved trajectory.

Paper: http://arxiv.org/abs/2302.02285

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Diffusion Inference (upper) and…
0
1
0

Fahim Farook

"Design Booster: A Text-Guided Diffusion Model for Image Translation with Spatial Layout Preservation. (arXiv:2302.02284v1 [cs.CV])" — A new approach for flexible image translation by learning a layout-aware image condition together with a text condition, which co-encodes images and text into a new domain during the training phase.

Paper: http://arxiv.org/abs/2302.02284

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Style image translation and sem…
0
1
0

Fahim Farook

"Divide and Compose with Score Based Generative Models. (arXiv:2302.02272v1 [cs.CV])" — Learning image components in an unsupervised manner in order to compose those components to generate and manipulate images in an informed manner.

Paper: http://arxiv.org/abs/2302.02272
Code: https://github.com/sandeshgh/Score-based-disentanglement

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Using all score components resu…
0
1
0

Fahim Farook

"Real-Time Image Demoireing on Mobile Devices. (arXiv:2302.02184v1 [cs.CV])" — A study on accelerating demoireing networks and a dynamic demoireing acceleration method (DDA) capable of real-time deployment on mobile devices.

Paper: http://arxiv.org/abs/2302.02184
Code: https://github.com/zyxxmu/DDA

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Images with moir ́e patterns. Th…
0
1
0

Fahim Farook

"DeepAstroUDA: Semi-Supervised Universal Domain Adaptation for Cross-Survey Galaxy Morphology Classification and Anomaly Detection. (arXiv:2302.02005v1 [astro-ph.GA])" — A universal domain adaptation method to overcome the challenge of artificial intelligence methods extracting dataset-specific, non-robust features.

Paper: http://arxiv.org/abs/2302.02005

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Example images from the source …
0
1
0

Fahim Farook

"PDEBENCH: An Extensive Benchmark for Scientific Machine Learning. (arXiv:2210.07182v4 [cs.LG] UPDATED)" — A benchmark suite of time-dependent simulation tasks based on Partial Differential Equations (PDEs), which comprises both code and data to benchmark the performance of novel machine learning models against both classical numerical simulations and machine learning baselines.

Paper: http://arxiv.org/abs/2210.07182
Code: https://github.com/pdebench/PDEBench

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
PDEBENCH provides multiple non-…
0
2
0

Fahim Farook

"Faster Attention Is What You Need: A Fast Self-Attention Neural Network Backbone Architecture for the Edge via Double-Condensing Attention Condensers. (arXiv:2208.06980v3 [cs.CV] UPDATED)" — A faster attention condenser design called double-condensing attention condensers that allow for highly condensed feature embeddings.

Paper: http://arxiv.org/abs/2208.06980

#AI #CV #NewPaper #DeepLearning #MachineLearning

<<Find this useful? Please boost so that others can benefit too 🙂>>
Proposed AttendNeXt architectur…
0
1
2
Show older