## Very Short Summary
The authors introduce a number of modifications to traditional hessian-free optimisation that makes the method work better for neural networks. The modifications are:
* Use the Generalised Gauss Newton Matrix (GGN) rather than the Hessian.
* Damp the GGN so that $G' = G + \lambda I$ and adjust $\lambda$ using levenberg-marquardt heuristic.
* Use an efficient recursion to calculate the GGN.
* Initialise each round of conjugated gradients with the final vector of the previous iteration.
* A new simpler termination criterion for CG. Terminate CG when the relative decrease in the objective falls below some threshold.
* Back-tracking of the CG solution. ie you store intermediate solutions to CG and only update if the new CG solution actually decreases the over all problem objective.

## Less Short Summary

### Hessian Free Optimisation in General
Hessian free optimisation is used when one wishes to optimise some objective $f(\theta)$ using second order methods but inversion or even computation of the Hessian is intractable or infeasible. The method is an iterative method and at each iteration, we take a second order approximation to the objective. i.e at iterantion n, we take a second order taylor expansion of $f$ to get:

$M^n(\theta) = f(\theta^n) + \nabla_{\theta}^Tf(\theta^n)(\theta - \theta^n) + (\theta - \theta^n)^TH(\theta - \theta^n) $

Where $H$ is the hessian matrix. If we minimise this second order approximation with respect to $\theta$ we would find that that $\theta^{n+1} = H^{-1}(-\nabla_{\theta}^Tf(\theta^n))$. However, inverting $H$ is usually not possible for even moderately sized neural networks.

There does however exist an efficient algorithm for calculating hessian vector products $Hv$ for any $v$. The insight of hessian-free optimisation is that one can solve linear problems of the form $Hx = v$ using only hessian vector products via the linear conjugated gradients algorithm. You therefore avoid the need to ever actually compute either the Hessian or its inverse.

To run vanilla hessian free all you need to do at each iteration is:

1) Calculate the gradient vector using standard backprop.

2) Calculate $H\theta$ product using an efficient recursion.

3) calculate the next update $\theta^{n+1} = ConjugatedGradients(H, -\nabla_{\theta}^Tf(\theta^n))$

The main contribution of this paper is to take the above algorithm and make the changes outlined in the very short summary.

## Take aways
Hessian-Free optimisation was perhaps the best method at the time of publication. Recently it seems that first order methods using per-parameter learning rates like ADAM or even learning-to-learn can outperform Hessian-Free. This is primarily because of the increased cost per iteration of Hessian Free. However it still seems that using curvature information if its available is beneficial though expensive.

More resent second order curvature appoximations like Kroeniker Factored Approximate Curvature (KFAC) and Kroeniker Factored Recursive Approximation (KFRA) are cheaper ways to achieve the same benefit.

#### Introduction

* Method to visualize high-dimensional data points in 2/3 dimensional space.
* Data visualization techniques like Chernoff faces and graph approaches just provide a representation and not an interpretation.
* Dimensionality reduction techniques fail to retain both local and global structure of the data simultaneously. For example, PCA and MDS are linear techniques and fail on data lying on a non-linear manifold.
* t-SNE approach converts data into a matrix of pairwise similarities and visualizes this matrix.
* Based on SNE (Stochastic Neighbor Embedding)
* [Link to paper](http://jmlr.csail.mit.edu/papers/volume9/vandermaaten08a/vandermaaten08a.pdf)

#### SNE

* Given a set of datapoints $x_1, x_2, ...x_n, p_{i|j}$ is the probability that $x_i$ would pick $x_j$ as its neighbor if neighbors were picked in proportion to their probability density under a Gaussian centered at $x_i$. Calculation of $\sigma_i$ is described later.
* Similarly, define $q_{i|j}$ as conditional probability corresponding to low-dimensional representations of $y_i$ and $y_j$ (corresponding to $x_i$ and $x_j$). The variance of Gaussian in this case is set to be $1/\sqrt{2}$
* Argument is that if $y_i$ and $y_j$ captures the similarity between $x_i$ and $x_j$, $p_{i|j}$ and $q_{i|j}$ should be equal. So objective function to be minimized is Kullback-Leibler (KL) Divergence measure for $P_i$ and $Q_i$, where $P_i$ ($Q_i$) represent conditional probability distribution given $x_i$ ($y_i$)
* Since KL Divergence is not symmetric, the objective function focuses on retaining the local structure.
* Users specify a value called perplexity (measure of effective number of neighbors). A binary search is performed to find $\sigma_i$ which produces the $P_i$ having same perplexity as specified by the user.
* Gradient Descent (with momentum) is used to minimize objective function and Gaussian noise is added in early stages to perform simulated annealing.

#### t-SNE (t-Distributed SNE)

##### Symmetric SNE

* A single KL Divergence between P (joint probability distribution in high-dimensional space) and Q (joint probability distribution in low-dimensional space) is minimized.
* Symmetric because $p_{i|j} = p_{j|i}$ and $q_{i|j} = q_{j|i}$
* More robust to outliers and has a simpler gradient expression.

##### Crowding Problem

* When we model a high-dimensional dataset in 2 (or 3) dimensions, it is difficult to segregate the nearby datapoints from moderately distant datapoints and gaps can not form between natural clusters.
* One way around this problem is to use UNI-SNE but optimization of the cost function, in that case, is difficult.

##### t-SNE

* Instead of Gaussian, use a heavy-tailed distribution (like Student-t distribution) to convert distances into probability scores in low dimensions. This way moderate distance in high-dimensional space can be modeled by larger distance in low-dimensional space.
* Student-t distribution is an infinite mixture of Gaussians and density for a point under this distribution can be computed very fast. 
* The cost function is easy to optimize.

##### Optimization Tricks

###### Early Compression

* At the start of optimization, force the datapoints (in low-dimensional space) to stay close together so that datapoints can easily move from one cluster to another. 
* Implemented an L2-penalty term proportional to the sum of the squared distance of datapoints from the origin.

###### Early Exaggeration

* Scale all the $p_{i|j}$'s so that large $q_{i|j}$*'s are obtained with the effect that natural clusters in the data form tight, widely separated clusters as a lot of empty space is created in the low-dimensional space.

##### t-SNE on large datasets

* Space and time complexity is quadratic in the number of datapoints so infeasible to apply on large datasets.
* Select a random subset of points (called landmark points) to display.
* for each landmark point, define a random walk starting at a landmark point and terminating at any other landmark point.
* $p_{i|j}$ is defined as fraction of random walks starting at $x_i$ and finishing at $x_j$ (both these points are landmark points). This way, $p_{i|j}$ is not sensitive to "short-circuits" in the graph (due to noisy data points).

#### Advantages of t-SNE

* Gaussian kernel employed by t-SNE (in high-dimensional) defines a soft border between the local and global structure of the data.
* Both nearby and distant pair of datapoints get equal importance in modeling the low-dimensional coordinates.
* The local neighborhood size of each datapoint is determined on the basis of the local density of the data.
* Random walk version of t-SNE takes care of "short-circuit" problem.

#### Limitations of t-SNE

* It is unclear t-SNE would perform on general **Dimensionality Reduction** for more than 3 dimensions. For such higher (than 3) dimensions, Student-t distribution with more degrees of freedom should be more appropriate.
* t-SNE reduces the dimensionality of data mainly based on local properties of the data which means it would fail in data which has intrinsically high dimensional structure (**curse of dimensionality**).
* The cost function for t-SNE is not convex requiring several optimization parameters to be chosen which would affect the constructed solution.

This paper argues that, in semi-supervised learning, it's suboptimal to use the same weight for all examples (as happens implicitly, when the unsupervised component of the loss for each example is just added together directly. Instead, it tries to learn weights for each specific data example, through a meta-learning-esque process. 

The form of semi-supervised learning being discussed here is label-based consistency loss, where a labeled image is augmented and run through the current version of the model, and the model is optimized to try to induce the same loss for the augmented image as the unaugmented one. The premise of the authors argument for learning per-example weights is that, ideally, you would enforce consistency loss less on examples where a model was unconfident in its label prediction for an unlabeled example. 

As a way to solve this, the authors suggest learning a vector of parameters - one for each example in the dataset - where element i in the vector is a weight for element i of the dataset, in the summed-up unsupervised loss. They do this via a two-step process, where first they optimize the parameters of the network given the example weights, and then the optimize the example weights themselves. To optimize example weights, they calculate a gradient of those weights on the post-training validation loss, which requires backpropogating through the optimization process (to determine how different weights might have produced a different gradient, which might in turn have produced better validation loss). This requires calculating the inverse Hessian (second derivative matrix of the loss), which is, generally speaking, a quite costly operation for huge-parameter nets. To lessen this cost, they pretend that only the final layer of weights in the network are being optimized, and so only calculate the Hessian with respect to those weights. They also try to minimize cost by only updating the example weights for the examples that were used during the previous update step, since, presumably those were the only ones we have enough information to upweight or downweight. With this model, the authors achieve modest improvements - performance comparable to or within-error-bounds better than the current state of the art, FixMatch. 

Overall, I find this paper a little baffling. It's just a crazy amount of effort to throw into something that is a minor improvement. A few issues I have with the approach: 

- They don't seem to have benchmarked against the simpler baseline of some inverse of using Dropout-estimated uncertainty as the weight on examples, which would, presumably, more directly capture the property of "is my model unsure of its prediction on this unlabeled example"
- If the presumed need for this is the lack of certainty of the model, that's a non-stationary problem that's going to change throughout the course of training, and so I'd worry that you're basically taking steps in the direction of a moving target
- Despite using techniques rooted in meta-learning, it doesn't seem like this models learns anything generalizable - it's learning index-based weights on specific examples, which doesn't give it anything useful it can do with some new data point it finds that it wasn't specifically trained on

Given that, I think I'd need to see a much stronger case for dramatic performance benefits for something like this to seem like it was worth the increase in complexity (not to mention computation, even with the optimized Hessian scheme)

[First off, full credit that this summary is essentially a distilled-for-my-own-understanding compression of Yannic Kilcher's excellent video on the topic] 

I'm interested in learning more about Neural Radiance Fields (or NERFs), a recent technique for learning a representation of a scene that lets you generate multiple views from it, and a paper referenced as a useful prerequisite for that technique was SIRENs, or Sinuisodial Representation Networks. In my view, the most complex part of understanding this technique isn't the technique itself, but the particularities of the problem being solved, and the ways it differs from a more traditional ML setup. 

Typically, the goal of machine learning is to learn a model that extracts and represents properties of a data distribution, and that can generalize to new examples drawn from that distribution. Instead, in this framing, a single network is being used to capture information about a single image, essentially creating a compressed representation of that image that brings with it some nice additional properties. 

Concretely, the neural network is representing a function that maps inputs of the form (x, y), representing coordinates within the image, to (r, g, b) values, representing the pixel values of the image at that coordinate. If you're able to train an optimal version of such a network, it would mean you have a continuous representation of the image. A good way to think about "continuous," here, is that, you could theoretically ask the model for the color value at pixel (3.5, 2.5), and, given that it's simply a numerical mapping, it could give you a prediction, even though in your discrete "sampling" of pixels, that pixel never appears.

Given this problem setting, the central technique proposed by SIRENs is to use sinusoidal non-linearities between the layers. On the face of it, this may seem like a pretty weird choice: non-linearities are generally monotonic, and a sine wave is absolutely not that. The appealing property of sinusoidal activations in this context is: if you take a derivative of a sine curve, what you get is a cosine curve (which is essentially a shifted sine curve), and the same is true in reverse. This means that you can take multiple derivatives of the learned function (where, again, "learned function" is your neural network optimized for this particular image), and have them still be networks of the same underlying format, with shifting constants. 

This allows SIRENs to use an enhanced version of what would be a typical training procedure for this setting. Simplistically, the way you'd go about training this kind of representation would be to simply give the inputs, and optimize against a loss function that reduced your prediction error in predicting the output values, or, in other words, the error on the f(x, y) function itself. When you have a model structure that makes it easy to take first and second derivatives of the function calculated by the model, you can, as this paper does, decide to train against a loss function of matching, not just the true f(x, y) function (again, the pixel values at coordinates), but also the first and second-derivatives (gradients and Laplacian) of the image at those coordinates. This supervision lets you learn a better underlying representation, since it enforces not just what comes "above the surface" at your sampled pixels, but the dynamics of the true function between those points. 

One interesting benefit of this procedure of using loss in a first or second derivative space (as pointed out in the paper), is that if you want to merge the interesting parts of multiple images, you can approximate that by training a SIREN on the sum of their gradients, since places where gradients are zero likely don't contain much contrast or interesting content (as an example: a constant color background). 

The Experiments section goes into a lot of specific applications in boundary-finding problems, which I understand at less depth, and thus won't try to explain. It also briefly mentions trying to learn a prior over the space of image functions (that is, a prior over the set of network weights that define the underlying function of an image); having such a prior is interesting in that it would theoretically let you sample both the implicit image function itself (from the prior), and then also points within that function.

DRL has lot of disadvantages like large data requirement, slow learning, difficult interpretation, difficult transfer, no causality, analogical reasoning done at a statistical level not at a abstract level etc. This can be overcome by adding a symbolic front end on top of DL layer before feeding it to RL agent. Symbolic front end gives advantage of smaller state space generalization, flexible predicate length and easier combination of predicate expressions. DL avoids manual creation of features unlike symbolic reasoning. Hence DL along with symbolic reasoning might be the way to progress for AGI. State space reduction in symbolic reasoning is carried out by using object interactions(object positions and object types) for state representation. Although certain assumptions are made in the process such as objects of same type behave similarly etc, one can better understand causal relations in terms of actions, object interactions and reward by using symbolic reasoning. 

Broadly, pipeline consists of (1)CNN layer - Raw pixels to representation (2)Salient pixel identification - Pixels that have activations in CNN above a certain threshold (3)Identify objects of similar kind by using activation spectra of salient pixels (4)Identify similar objects in consecutive time steps to track object motion using spatial closeness(as objects can move only by a small distance in consecutive frames) and similar neighbors(different type of objects can be placed close to each other and spatial closeness alone cannot identify similar objects) (4)Building symbolic interactions by using relative object positions for all pairs of objects located within a certain maximal distance. Relative object position is necessary to capture object dynamics. Maximal distance threshold is required to make the learning quicker eventhough it may reach a locally optimal policy (4)RL agent uses object interactions as states in Q-Learning update. Instead of using all object interactions in a frame as one state, number of states are further reduced by considering interactions between two types to be independent of other types and doing a Q-Learning update separately for each type pair. Intuitive explanation for doing so is to look at a frame as a set of independent object type interactions. Action choice at a state is then the one that maximizes sum of Q values across all type pairs. 

Results claim that using DRL with symbolic reasoning, transfer in policies can be observed by first training on evenly spaced grid world and using it for randomly spaced grid world with a performance close to 70% contrary to DQN that achieves 50% even after training for 1000 epochs with epoch length of 100.