Today's text-entry tools offer a plethora of interface technologies to support users in a variety of situations and with a range of different input methods and devices.16 Recent hardware developments have enabled remarkable innovations, such as virtual keyboards that allow users to type in thin air, or to use their body as a surface for text entry. Similarly, advances in machine learning and natural language processing have enabled high-quality text generation for various purposes, such as summarizing, expanding, and co-authoring. As these technologies rapidly develop, there has been a rush to incorporate them into existing systems, often with little thought for the interactivity problems this may cause. The use of large language models (LLMs) to speed up text generation and improve prediction or completion models is becoming increasingly commonplace, with enormous theoretical efficiency savings;29 however, the implementation of these LLMs into text-entry interfaces is crucial to realizing their potential.

Originality: The authors propose a novel relaxation (to the best of my knowledge) for networks with ReLU activations that tighten previously proposed relaxations that ignore the correlations between neurons in the network. The theoretical results are also novel (although unsurprising). However, it would be useful for the authors to better clarify the computational requirements and tightness of k-ReLU relative to DeepPoly and other similar relaxations and bound propagation methods like [13] and https://arxiv.org/abs/1805.12514, Quality: The theoretical results are accurate (albeit unsurprising) in my opinion. The experimental section is missing several important details in my opinion: 1) The authors say that experiments are performed on both MNIST and CIFAR-10, but the tables 2/3 only report numbers on MNIST.

This paper presents experiments with shallow network trained to mimic the predictions of deeper, more complex networks trained on the same data. When trained to predict the pre-softmax outputs of the "teacher" networks, these mimic networks are able to approximate the performance of the deeper networks, albeit with more parameters. They consistently outperform networks with the same architecture trained the cross-entropy objective function. The paper presents an interesting -- and perhaps useful -- set of experiments, but the theoretical analysis is extremely weak. There are a number of bold claims that in my opinion are not demonstrated by the experimental work, and many speculative statements, the truth of which there has been little effort to investigate.

This is a relevant and timely paper that has been reviewed by four knowledgeable referees, who also thoroughly considered the author's response to their initial reviews. Three of these reviewers recommend acceptance, providing detailed suggestions on how to improve this work before its final submission. This dissenting opinion was upheld by R3 after discussion with other referees. R3 in my opinion correctly brings up that if the proposed approach aims to improve runtime with an approximate algorithm, this must be sufficiently demonstrated in experiments vs. straightforward alternatives (such as retraining-based methods). That has not been done in the original submission neither in the rebuttal.

Weaknesses: - The authors compared their method to the baseline approach only. However, there are plenty of curriculum learning methods that could have been used as relevant state-of-the-art competing methods to compare with, e.g. Comparison with such competing methods is mandatory, in my opinion. I believe that the non-linearity is typically applied before the pooling operation. Even so, it is not clear why it works so well.

In this paper, the authors propose a distributed Newton method for gradient-norm optimization. The method does not impose any specific form on the underlying objective function. The authors present convergence analysis for the method and illustrate the performance of the method on a convex problem (in the main paper). Originality: The topic of the paper, in my opinion, is very interesting. The paper presents an efficient Newton method that is motivated via the optimization of the norm of the gradient.

Originality: To the best of my knowledge the model of general user retention dynamics and corresponding statements evidencing negative feedback loops are novel contributions to the literature in sequential fairness works. The contributions of the paper would be clearer if citations were provided for methods and models introduced in earlier works (for example, I suggest adding citations for the fairness criteria in lines 149-158, for user departure models in 197-208, and for the statement in lines 173-174, if applicable). Since the full related work is deferred to the appendix, I see no need to cite [2, 3, 7, 10, 15, 16] without distinction between them. More context on what these works do and how they relate to your work is useful for readers to contextualize your contributions; please expand on the discussion of these papers. Quality: The simple and unifying model of sequential decision making presented is very valuable in my opinion.

There was a ton of discussion about this paper between reviewers and area chairs, multiple reviewers improved their view of the paper based on the author response, and I read through the paper in detail myself. I was still conflicted after reading it, but I am leaning towards recommending acceptance. However, I *implore* the authors to carefully consider the issues brought up by R2 and R3 as well as the issues that I bring up below. I believe that every single one of the issues brought up can be fixed, and will be extremely-disappointed if these issues are not addressed in the final version (it would make me regret recommending acceptance, and probably be more-harsh on empirical papers in the future, especially by the same authors). Some specific comments from my read-through of the paper: - I agree with the authors that optimization is a mix of theory and empirical work, and it is completely ok for works to be purely empirical if the experiments are done well.

Generating images from brain waves is gaining increasing attention due to its potential to advance brain-computer interface (BCI) systems by understanding how brain signals encode visual cues. Most of the literature has focused on fMRI-to-Image tasks as fMRI is characterized by high spatial resolution. However, fMRI is an expensive neuroimaging modality and does not allow for real-time BCI. On the other hand, electroencephalography (EEG) is a low-cost, non-invasive, and portable neuroimaging technique, making it an attractive option for future real-time applications. Nevertheless, EEG presents inherent challenges due to its low spatial resolution and susceptibility to noise and artifacts, which makes generating images from EEG more difficult. In this paper, we address these problems with a streamlined framework based on the ControlNet adapter for conditioning a latent diffusion model (LDM) through EEG signals. We conduct experiments and ablation studies on popular benchmarks to demonstrate that the proposed method beats other state-of-the-art models. Unlike these methods, which often require extensive preprocessing, pretraining, different losses, and captioning models, our approach is efficient and straightforward, requiring only minimal preprocessing and a few components. The code is available at https://github.com/LuigiSigillo/GWIT.

In recent years, toxic content and hate speech have become widespread phenomena on the internet. Moderators of online newspapers and forums are now required, partly due to legal regulations, to carefully review and, if necessary, delete reader comments. This is a labor-intensive process. Some providers of large language models already offer solutions for automated hate speech detection or the identification of toxic content. These include GPT-4o from OpenAI, Jigsaw's (Google) Perspective API, and OpenAI's Moderation API. Based on the selected German test dataset HOCON34k, which was specifically created for developing tools to detect hate speech in reader comments of online newspapers, these solutions are compared with each other and against the HOCON34k baseline. The test dataset contains 1,592 annotated text samples. For GPT-4o, three different promptings are used, employing a Zero-Shot, One-Shot, and Few-Shot approach. The results of the experiments demonstrate that GPT-4o outperforms both the Perspective API and the Moderation API, and exceeds the HOCON34k baseline by approximately 5 percentage points, as measured by a combined metric of MCC and F2-score.