at least something worth reading

mllm arch, image-> patchify -> qformer -> condition llm.

stupid majority voting

agent installs env, tools etc...

why store the mahcine state? if the agent weights are constant? not sure.... conversation state, repo state still make sense.

decompose task into smaller ones, and allocate then, ,

scaling agent systems.

interesting, so they don't need agents. but if the agents are robots, this is needed, its a physical constraint.

basically, central agent doing reasoning and planning > multi-agent,

however, master can assign tasks to several slaves, they can asynchronously perform them and return.

toolformer

one dude can reward hack itself so instead have two dudes one dude does solution other dude critiques it and they continue until the dude critiquiing it is fooled.

suttons actor-critic model.

4 kinds of agentic patterns: - reflection - tool use - planning - multi-agent collaboration and coordination

agent has 4 things: - interact with user - call tool - access memory as rag - some sort of planner component.

so true

in data parallelism : master gpu - weights copied to slave gpus - fwd pass done on part of data - gradients sent back to master, summed and master model updated. - so for every feed-forward, every model on every slave device is essentially destroyed, makes it slow.

here different layers of a model go to different gpus.

what is direct set prediction problem?

sentence embedding. not word embedding

three types of multi task models are there.

what is the difference between transfer learning and multi task learning.

somewhat similar to ohem.

Methods of evaluating similarity ty

either you can make image clearer or you can vary it. This was the conventional wisdom.

Annotated by @rajatmodi62

Need to understand what is wassertein loss. Will implement both of these.

hello

They have improved the information flow in the network of multiple masksts

If we use large learning rates, we cannot use that much regularization.

learning rate test-> plot a graph of test accuracy with learning rate. Ideally, the learning rate appears to overfit or underfit. Training only starts if the learning rate is kept to be small. We can increase it later.

If large region in Lr test is flat, we can take that range of learning rates and obtain super convergence

Takeaway , plot the test loss for multiple learning rates. Lower learning rate will show that test loss will continue to decrease. It does not increase anywhere so underfitting.

Increase the lr then , and see that initial falling is there, after which the accuracy does not improve.

old papers kept large learning rates and small batch size

Old R-CNN, pipeline region extraction->warp in a single image->feed to cnn->o/p is bounding box coordinate and class,classification is done by passing to a family of svm trained on a classification task .

Needs 2000 regions, slow since the proposals need to be deesgined for each of the region

They don't rely on image optimization for style transfer. Instead they are training feed forward networks to apply style to an input image. I need to look at the code to understand can they apply the style in real time. _> in a single forward pass. The loss comes from the vgg model , maybe they use generator- discrominator architecture for the dcgan work , check it later on (supplementary material)

Then what is hte number of iterations which are present in the table.

loss functions from high features

Image annotation methods are used

[Future Idea]

Zero Padding can impact object recognition performance. Some neurons get trained to look at the zero paddinf and ht eimage edge,

A girl is there, she is repeated across entire image. In first layer output many girls repeated, ideally in the end layer it should represent just one girl in the image. In CNN, at higher layers, only filters remain , size of the activations reduce. Take the motivation from there to understand the result.

[Look at the texture information] Texture refers to the localization of thepattern, ideally the pattern should not be repeated across the entire image. That's why they multiplied the feature activations, to lose the spatial information .

[In the results], the outputs of lower layers repeat the patterns throughout the image. In the higher layer output , the locality of the pattern is there. Only once unique girl is seen, image information appears to be reflected properly.

[Max pool->Average Pool] Layers replaced to get better statistics,

This is a test annotation

• CNN only predicts if pneumothorax there or not
• FCN uses UNET with attention. It requires labels at a pixel level. However, they have a notion of classification of "area" of an selected image. -MIL has the notion of using the image classification labels for local prediction. Image divided into patches (448X448) . For each patch, the global label is seen .

If pneuomothorax is not there , then none of the patch cntains the disease Otherwise, one of the patch at least has the disesase , and the best patch is tried to be found out.

-Disadvantage- MIL only offers the rectangular patches information as of now .

CNN are good for classification problems. When local information is needed, like where exactly the pneumothorax is, they face a problem.

basic

Uses CNN. formulates a style function from later layers of the cnn , and the overall image function from the earlier layers of the cnn

Completed CycleGAN , use lateron

Collection of the Neural Style Transfer Papers

checking the annotation tool in the check box