Rajat Modi

Talks

ComputerVision Talks: Asynchronous Perception Machine
Explaining how APMs were a step towards getting GLOM to work well.

Making GLOM Work
GLOM was an idea by Hinton et al in 2021. Here, we discuss what were some key ideas towards getting it to work.

Putting an End to End to End Gradient Descent
A learning algorithm which trains each layer of a neural net in ‘isolation’ greedily. Optimization is based on a local-constraint (called Greedy-InfoMax)

Forward Forward Algorithm: Part 1/2
Neural net is optimized by piping input data sequentially with two forward passes

Forward Forward Algorithm: Part 2/2
Discussions on philosophical ideas in Forward Forward paper, for eg, getting rid of gradient descent, mortal computation.

Biologically Plausible Learning

Backpropagation And Brain
How can backpropagation be made more biologically plausible?

Target Propagation
Dr. Bengio's approach to avoid computing derivatives of neurons. Pipe error signal back into the network, output of each neuron is the actual derivative

Autoencoding Assumption
Autoencoder can be used to compute top-down neural activities on the fly. Overcomes backprop's bottleneck to store intermediate representations

Weight Transport Problem
Backpropagation requires computing transpose of weight matrix to calculate error signal. Brain doesn't do transpose.

Memory Explosion Issue in Attention
Attention is quadratic in time/memory. Linear Attention/Mamba etc can do linear time/constant memory. Linear time is still inefficient.

[Cognitive Side]
Necker Cube / Superposition
Mount Everest

Representation Supersedes Scaling.

It does not mean that we should not scale AI. It just means that we should scale it intelligently. Bruteforce scaling wastes compute. A good representation can help scaling models find optimal solution faster. Hence, representations are still important.

Some Interesting Stuff

Artificial Karamuto Oscillatory Neurons
AKORN (similar to a seed) is Takeru Miyato's idea in 2025. Meant as a replacement for Mcullough Pitts Neuron.

Philosophy Of Mind

On Reflections of a Mortal Machine

The nature of soul in a learning machine.

On the higher purpose behind research

On building emotions into machines.

On the destructiveness and pain in machines.