Datasets:
anonymousatom
/
SFT15KQ_OCR

Dataset card Data Studio Files
xet
 Community
1
Paper_ID
stringlengths
10
10
Question
stringlengths
201
1.81k
ocr_output
stringlengths
252
54k
⌀
rEQ8OiBxbZ
Could you elaborate on how the local structures are reconstructed? What serves as the input for this process: a single embedding from the TokenGT-3D output, or a collection of embeddings from local structure segmentations within a single molecule?
3D Molecular Pretraining via Localized Geometric Generation Anonymous authors Paper under double-blind review Abstract Self-supervised learning on 3D molecular structures has gained prominence in AI-driven drug discovery due to the high cost of annotating biochemical data. However, few have studied the selection of ...
s6bKLlF4Pe
I am doubtful about the significance of convergence results. The convergence result with GPI follows the same rate as the convergence rate without GPI. It is hard to tell directly what is the difference in the constants. Having a thorough discussion with some examples would serve to give readers a better understanding ...
Proviable Knowledge Transfer using Successor Feature for Deep Reinforcement Learning Anonymous authors Paper under double-blind review Abstract This paper studies the transfer reinforcement learning (RL) problem where multiple RL problems have different reward functions but share the same underlying transition dynam...
xC8xh2RSs2
The paper uses exact keyword matching to identify corresponding subsections. Thus it's hard to know the proportion of dataset cards which covers the corresponding subsection but with different keywords.
Navigating Dataset Documentations in AI: A Large-Scale Analysis of Dataset Cards on Hugging Face Xinyu Yang * Cornell University xy468@cornell.edu Weixin Liang* Stanford University wxliang@stanford.edu James Zou Stanford University jamesz@stanford.edu Abstract Advances in machine learning are closely tied to the c...
Kz3yckpCN5
The implicit assumption of this work (revealed in the title) is that there exists a claim or understanding that imitating proprietary language models by sampling their outputs for training is all that is needed to achieve performance parity - however, I contend that this isn't the prevalent understanding.
THE FALSE PROMISE OF IMITATING PROPRIETARY LANGUAGE MODELS Arnav Gudibande*, Eric Wallace*, Charlie Snell* Xinyang Geng, Hao Liu, Pieter Abbeel, Sergey Levine, Dawn Song UC Berkeley {arnavg, ericwallace, csnell22}@berkeley.edu ABSTRACT An emerging method to cheaply improve a weaker language model is to finetune it o...
dBO8ZPQMVF
Can you help me understand better the relationship between MAS and a standard diffusion model? Can MDM be seen as the forward process, why the 3D consistency check as the reverse process? Or is this incorrect?
MAS: Multi-view Ancestral Sampling for 3D Motion Generation Using 2D Diffusion Anonymous authors Paper under double-blind review Abstract We introduce Multi-view Ancestral Sampling (MAS), a method for generating consistent multi-view 2D samples of a motion sequence, enabling the creation of its corresponding 3D coun...
TYXtXLYHpR
In the related works, you distinguish your method from shapelet-based methods, stating that these are primarily used for data mining and classification tasks. However, if these shapelet methods are unsupervised (e.g., Karlsson, Isak, Panagiotis Papapetrou, and Henrik Boström.
Towards Transparent Time Series Forecasting Krzysztof Kacprzyk University of Cambridge kk751@cam.ac.uk Tennison Liu University of Cambridge t1522@cam.ac.uk Mihaela van der Schaar University of Cambridge The Alan Turing Institute mv472@cam.ac.uk Abstract Transparent machine learning (ML) models are es...
FvK2noilxT
In Sec 4.1 Training dataset, why did the authors use different standard deviations to noise the MANO parameters for translation, rotation, and pose parameters Does the way to noise the training sets affect the learning?
GENEOH DIFFUSION: TOWARDS GENERALIZABLE HAND-OBJECT INTERACTION DENOISING VIA DENOISING DIFFUSION Xueyi Liu1,3 Li Yi1,2,3 1Tsinghua University 2Shanghai AI Laboratory 3Shanghai Qi Zhi Institute Project website: meowuu7.github.io/GeneOH-Diffusion ABSTRACT In this work, we tackle the challenging problem of denoising h...
otHZ8JAIgh
Since both PID and PIB rely on sampling from distribution, it does seem that the performance will indeed by affected by which samples are chosen or how many of them are sampled. The discussion around this point needs to be made explicit.
Prototypical Information Bottlenecking and Disentangling for Multimodal Cancer Survival Prediction Yilan Zhang \(^1\)\(^2\)\(^1\), Yingxue Xu \(^1\)\(^1\), Jianqi Chen \(^2\), Fengying Xie \(^*\)\(^2\), Hao Chen \(^*\)\(^1\) \(^1\)The Hong Kong University of Science and Technology, \(^2\)Beihang University yxueb@con...
9NKRfhKgzI
I'm having difficulty understanding the reason for conditioning on u for formalizing Goal 2 and Goal 3, as the first goal's objective already minimizes the I(z,u), why does the method need to maximize/minimize the conditional mutual information instead of just the mutual information?
Adversarially Robust and Privacy-Preserving Representation Learning via Information Theory Anonymous authors Paper under double-blind review Abstract Machine learning models are vulnerable to both security attacks (e.g., adversarial examples) and privacy attacks (e.g., private attribute inference). Existing defenses...
Uj2Wjv0pMY
Since this paper is about the new dataset which is claimed to focus on error recognition, however there not much new insights about the significance of bringing more error videos to procedural video dataset: neither in the way data is captured or significant baselines, experiments to demonstrate why it matters?
Put on your detective hat: What’s wrong in this video? A Dataset for Error Recognition in Procedure Videos Anonymous authors Paper under double-blind review Abstract Following step-by-step procedures is an essential component of various activities carried out by individuals in their everyday lives. These procedures ...
TLADT8Wrhn
The fundamental issue of continual learning is catastrophic forgetting. If we fine-tune a small number of parameters (e.g., prompt tuning) in the CLIP model, is catastrophic forgetting a major concern? On the other hand, if we fine-tune a large number of parameters, resource limitations may become a factor. Therefore, ...
TiC-CLIP: Continual Training of CLIP Models Saurabh Garg†∗ Mehrdad Farajtabar† Hadi Pouransari† Raviteja Vemulapalli† Sachin Mehta† Oncel Tuzel† Vaishaal Shankar† Fartash Faghri† †Apple ‡Carnegie Mellon University sgarg2@andrew.cmu.edu, fartash@apple.com Abstract Keeping large foundation models up to date on latest...
KPmajBxEaF
Does the proposed method work well with different frame numbers during inference time? As the proposed 2d-3d mapping needs aggregating information from all frames, does it consume huge memory in dense views (e.g., more than 100 frames)?
LEAP: Liberate Sparse-View 3D Modeling from Camera Poses Hanwen Jiang Zhenyu Jiang Yue Zhao Qixing Huang Department of Computer Sciences, University of Texas at Austin Project page: https://hwjiang1510.github.io/LEAP/ Figure 1: LEAP performs 3D modeling from sparse views without camera pose information. We s...
PFdjJiZjPj
In the “All-generated” setting, each problem is associated with multiple solutions. Do you give all of these simultaneously to the model and ask it to generate tests? Or one at a time, with one set of tests per solution?
THE PROGRAM TESTING ABILITY OF LARGE LANGUAGE MODELS FOR CODE Anonymous authors Paper under double-blind review ABSTRACT Recent development of large language models (LLMs) for code like CodeX and CodeT5+ demonstrates tremendous promise in achieving code intelligence. Their ability of synthesizing code that completes...
t5LXyWbs5p
The experimental setting of unimodal vs multimodal is very confusing. The authors state that ExpEMG is a dataset of single-channel EMG recordings, how is the multimodal approach applied in this case? This same problem also applies to other datasets.
Frequency-Aware Masked Autoencoders for Multimodal Pretraining on Biosignals Anonymous authors Paper under double-blind review Abstract Leveraging multimodal information from biosignals is vital for building a comprehensive representation of people’s physical and mental states. However, multimodal biosignals often e...
YnaGcMJQ0M
One analogy is the subset of OOD methods that try to fit a density to the test points to do likelihood ratio tests (the density is evaluated on each test datapoint but is the result of learning on all test points). If I understand correctly, though you briefly acknowledge using geometry of the whole test set as a motiv...
Detecting Out-of-Distribution Samples via Conditional Distribution Entropy with Optimal Transport Anonymous authors Paper under double-blind review Abstract When deploying a trained machine learning model in the real world, it is inevitable to receive inputs from out-of-distribution (OOD) sources. For instance, in c...
6pPYRXKPpw
What is the action space used by the environments? In the diffusion policy paper they show that diffusion policies are better for some absolute action spaces while being worse for others relative action spaces. Clarification as to that would be great.
Towards Diverse Behaviors: A Benchmark for Imitation Learning with Human Demonstrations Xiaogang Jia∗†‡ Denis Blessing† Xinkai Jiang†‡ Moritz Reuss‡ Atalay Donat† Rudolf Lioutikov† Gerhard Neumann† † Autonomous Learning Robots, Karlsruhe Institute of Technology ‡ Intuitive Robots Lab, Karlsruhe Institute of Technology...
Jg8y1buQ3r
Can one gain any explainability with regards to the memory module? What does it actually learn? It seems like a black box that has been named memory module and untenably attributed with correlation-extracting functionality.
LLM-driven Hateful Meme Detection via Cross-modal Memorizing and Self-rejection Training Anonymous authors Paper under double-blind review Abstract Hateful meme detection (HMD) is critical for determining whether online multimodal content carries harmful information, which plays a pivotal role in maintaining a harmo...
XUCAA0XnPC
Even if it does, how can the client know if the server is malicious in its part of the models as well. For instance, it could shuffle or modify the outputs of its server nets such that only some of them are useful for the client model.
ENSEMBLER: COMBATING MODEL INVERSION ATTACKS USING MODEL ENSEMBLE DURING COLLABORATIVE INFERENCE Anonymous authors Paper under double-blind review ABSTRACT Deep learning models have exhibited remarkable performance across various domains. Nevertheless, the burgeoning model sizes compel edge devices to offload a sign...
PczQtTsTIX
At the end of the introduction, you say that your success with batch norm “contradicts” another paper [2] that did not find batch norm to work well. Why do you think you were able to achieve better results? Is it because you removed the target network, or is there another reason?
CrossQ: Batch Normalization in Deep Reinforcement Learning for Greater Sample Efficiency and Simplicity Aditya Bhatt*1,4 Daniel Palenicek*1,2 Boris Belousov1,4 Max Argus3 Artemij Amiranashvili3 Thomas Brox3 Jan Peters1,2,4,5 *Equal contribution 1Intelligent Autonomous Systems, TU Darmstadt 2Hessian.AI 3University of F...
3y2TfP966N
The task in Sec 4.2.1 aims to regress difference in the representation to the time distance. As stated in point 1, again, various pattern difference might be contained in the representations from arbitrary pairs of time series, therefore it's very likely that they can not regress to the consistent time distance. Same c...
T-Rep: Representation Learning for Time Series using Time-Embeddings Archibald Fraikin Let it Care PariSanté Campus, Paris, France archibald.fraikin@inria.fr Adrien Bennetot Let it Care PariSanté Campus, Paris, France adrien.bennetot@letitcare.com Stéphanie Allassonière Université Paris Cité, INRIA, In...
ZZTkLDRmkg
Furthermore, there is some curiosity regarding whether the results of the baseline methods were adequately compared. It appears from Appendix F that the internal and boundary grids for the baseline methods are distinguished using one-hot encoding. It's worth considering if this method provides the fairest basis for com...
BENO: Boundary-Embedded Neural Operators for Elliptic PDEs Haixin Wang\textsuperscript{1,*}, Jiaxin Li\textsuperscript{2,*}, Anubhav Dwivedi\textsuperscript{3}, Kentaro Hara\textsuperscript{3}, Tailin Wu\textsuperscript{2,†} \textsuperscript{1}National Engineering Research Center for Software Engineering, Peking Unive...
ikwEDva1JZ
What happens when the representation function is of a different form? If either the transformer does not have enough layers or the width, is there an approximate representation function learned on which regression is performed, or does the entire mechanism fall apart?
How Do Transformers Learn In-Context Beyond Simple Functions? A Case Study on Learning with Representations Tianyu Guo¹ Wei Hu² Song Mei¹ Huan Wang³ Caiming Xiong³ Silvio Savarese³ Yu Bai³ ¹UC Berkeley ²University of Michigan ³Salesforce AI Research tianyu_guo@berkeley.edu Abstract While large language models based ...
s25i99RTCg
However, in a scenario where inference is desired based solely on modality A to predict B, would a masked C still be necessitated? This raises questions about the model's flexibility and its adaptability to accommodate various generative scenarios with different modalities. The capacity to dynamically adjust to these c...
ABSTRACT Multi-modal data-sets are ubiquitous in modern applications, and multi-modal Variational Autoencoders are a popular family of models that aim to learn a joint representation of the different modalities. However, existing approaches suffer from a coherence–quality tradeoff, where models with good generation qu...
u3RJbzzBZj
The paper mentions that “The Placeholder technique shares similarities with the currently popular Masking technique in the unsupervised pretraining domain”, however, this claim requires further explanation to strengthen its validity.
null
m2NVG4Htxs
The pass rate is significantly lower for easy and medium problems, even for log(Github Presence) = 0. I understand that GitHub Presence is a proxy, but I would think that log(GitHub Presence) = 0 is our best guess for
TO THE CUTOFF... AND BEYOND? A LONGITUDINAL PERSPECTIVE ON LLM DATA CONTAMINATION Manley Roberts¹, Himanshu Thakur¹,², Christine Herlihy³, Colin White¹, Samuel Dooley¹ ¹Abacus.AI ²Carnegie Mellon University ³University of Maryland {manley,colin,samuel}@abacus.ai; hthakur@andrew.cmu.edu; cherlihy@umd.edu ABSTRACT Rec...
sAOtKKHh1i
In Table 1, are the antmaze results obtained by conditioning on the state or on images observations? I suspect this is the state. If I'm right, how did you adapt SSP and SFP, which are designed to work with images? Isn't this comparison unfair, since SSP and SFP, which are designed to work with images?
Subwords as Skills: Tokenization for Sparse-Reward Reinforcement Learning Anonymous authors Paper under double-blind review Figure 1: A sample of some “skills” that our method identifies for the (a) AntMaze and (b) Kitchen environments, where the transparency is higher (color is paler) for poses earlier in the trajec...
yAcLwJu9qs
Humans are presented with one image at a time for 200 ms? Isn’t it too short to notice the image? Are the human participants in an average recognize objects in the image within that time? Would it be safe to assume that human participants do even better job when presented with an image upto 1s? It is mentioned that tim...
ASSESSING VISUALLY-CONTINUOUS CORRUPTION ROBUSTNESS OF NEURAL NETWORKS RELATIVE TO HUMAN PERFORMANCE Anonymous authors Paper under double-blind review ABSTRACT While Neural Networks (NNs) have surpassed human accuracy in image classification on ImageNet, they often lack robustness against image corruption, i.e., cor...
ikdB0VXPlw
Regarding the experiment, the proposed method archives competitive results on KIT in terms of FID, but not other metrics or on HumanML3D. It will be helpful to discuss why is this the case. It is perfectly fine to not attain SOTA on everything, but studying the limitations can provide the community with key insights.
Motion Flow Matching for Efficient Human Motion Synthesis and Editing Anonymous authors Paper under double-blind review Abstract Human motion synthesis is a fundamental task in the field of computer animation. Recent methods based on diffusion models or GPT structure demonstrate commendable performance but exhibit d...
s6X3s3rBPW
Or, if we are assessing Subject Knowledge, can't that be done by MCQA, which doesn't require expert annotation once the benchmark is created? If we are assessing programming, can't we check that with pre-specified unit tests? Indeed, the appendix shows that the unidentifiable datasets used in this paper are multiple ch...
Efficiently Measuring the Cognitive Ability of LLMs: An Adaptive Testing Perspective Anonymous authors Paper under double-blind review Abstract Large language models (LLMs), like ChatGPT, have shown human-level cognitive ability. Benchmarks from various fields (e.g., Literature, Biology and Psychology) are often use...
BifeBRhikU
Inconsistent Salient Weight Methodology between PTQ and QAT: The absence of a consistent methodology for salient weight protection between PTQ and QAT is concerning. While the effectiveness of using Hessian criteria for identifying salient weights in PTQ is demonstrated through performance comparisons, the rationale fo...
PB-LLM: Partially Binarized Large Language Models Zhihang Yuan Houmo AI Zhen Dong UC Berkeley Abstract This paper explores network binarization, a radical form of quantization, compressing model weights to a single bit, specifically for Large Language Models (LLMs) compression. Due to previous binarization meth...
9Cu8MRmhq2
While the authors effectively illustrate the motivation in Fig. 1, the advantages of the proposed OT method over DTW require more elaboration. It is advisable to include further discussions to expound upon and clarify the claims made regarding the superiority of OT over DTW.
MULTI-GRANULARITY CORRESPONDENCE LEARNING FROM LONG-TERM NOISY VIDEOS Yijie Lin\textsuperscript{1} Jie Zhang\textsuperscript{2} Zhenyu Huang\textsuperscript{1} Jia Liu\textsuperscript{1} Zujie Wen\textsuperscript{3} Xi Peng\textsuperscript{1,*} \textsuperscript{1}Sichuan University \textsuperscript{2}Beijing Universi...
KkrDUGIASk
The paper mentioned new agent privacy issue. I assume the late participate will require the new agent to access the fused feature to do the update according to equations in Section 4.3. Will the fused feature release some privacy of the old agents to the new agent?
AN EXTENSIBLE FRAMEWORK FOR OPEN HETEROGENEOUS COLLABORATIVE PERCEPTION Yifan Lu\textsuperscript{1,4}, Yue Hu\textsuperscript{1,4}, Yiqi Zhong\textsuperscript{2}, Dequan Wang\textsuperscript{1,3}, Yanfeng Wang\textsuperscript{1,3}, Siheng Chen\textsuperscript{1,3,4}\textsuperscript{✉}, \textsuperscript{1} Shanghai Jia...
RDSj6S8WJe
Hierarchical Structures in Real-world Scenarios: With the proposed dynamics aggregation framework depending heavily on the hierarchical structure of problems, how feasible is it to identify or establish such hierarchies in complex, real-world scenarios, where the state dynamics might be more intricate and less structur...
DEMYSTIFYING LINEAR MDPs AND NOVEL DYNAMICS AGGREGATION FRAMEWORK Joongkyu Lee Graduate School of Data Science Seoul National University jklee0717@snu.ac.kr Min-hwan Oh Graduate School of Data Science Seoul National University minoh@snu.ac.kr ABSTRACT In this work, we prove that, in linear MDPs, the feature dimensi...
PuCno7nwgH
When building the hyper edges for the proposed model, the authors used secondary interaction between categorical features. What's the time and storage complexity for the proposed algorithm? Will it explode the system if there are a lot of categorical features available for the users and items?
Categorical Entity Features in Recommendation Systems Using Graph Neural Networks Anonymous authors Paper under double-blind review Abstract Graph neural networks are widely used in recommender engines and are commonly applied to user-item graphs augmented by various side information, including categorical entity fe...
3NXhwkZGjz
The proposed pseudo label consolidation method seems similar to HCL Huang et al. (2021) which also aggregates predictions from multiple models/hypotheses to regularize/generate the final pseudo labels. Please discuss the differences, advantages and disadvantages of HCL and the proposed method.
Source-Free Unsupervised Domain Adaptation with Hypothesis Consolidation of Prediction Rationale Anonymous authors Paper under double-blind review Abstract Source-Free Unsupervised Domain Adaptation (SFUDA) is a challenging task where a model needs to be adapted to a new domain without access to target domain labels...
QHzzAU7Qf9
The presented method is a weighted average of parameters. How does such a method regularise against particularly known failure modes of MoEs which often require explicit regularisation such as through load, importance, entropy, or Mutual Information?
SOFT MERGING OF EXPERTS WITH ADAPTIVE ROUTING Anonymous authors Paper under double-blind review ABSTRACT Neural networks that learn to route their inputs through different “expert” subnetworks provide a form of modularity that standard dense models lack. Despite their possible benefits, modular models with learned r...
TW0MVSflg5
The authors use a 2D-CNN to obtain super-pixel features, meaning that the features for similarity computation are derived from a pixel region, not individual pixels. This approach seems misaligned with the NeRF setting, which relies only on per-pixel information instead of per-region data. Such a discrepancy makes the ...
SELF-EVOLVING NEURAL RADIANCE FIELDS Anonymous authors Paper under double-blind review ABSTRACT Recently, neural radiance field (NeRF) has shown remarkable performance in novel view synthesis and 3D reconstruction. However, it still requires abundant high-quality images, limiting its applicability in real-world scen...
8Itp6Axs9Z
More importantly, the experimental results do not seem to support the central thesis of the paper, namely that the SelfDreamer technique is beneficial specifically in the frame-masked setting: * SelfDreamer performs better than the alternatives from the very beginning (without frame-masking). * In the experiment with 2...
SelfDreamer: Dual-Prototypical Regularization for Frame-Masked Model-Based Reinforcement Learning Anonymous authors Paper under double-blind review Abstract In the realm of reinforcement learning (RL), the conventional approach involves training agents in unknown environments using extensive experiences comprising h...
bJ3gFiwRgi
While Equation 12 (Appendix A.1) is the dual of Equation 11, if the domain is extended from linear constraint functions to non-linear constraint functions, the equation would no longer behave as the dual of the original problem as formulated in Equation 11, right? Does it make sense to use this as the lower level probl...
Meta Inverse Constrained Reinforcement Learning: Convergence Guarantee and Generalization Analysis Shicheng Liu & Minghui Zhu Department of Electrical Engineering Pennsylvania State University University Park, PA 16802, USA {sfl5539,muz16}@psu.edu Abstract This paper considers the problem of learning the reward func...
Dxl0EuFjlf
Similar to question 1, in many cases, the amplitude shifting loss will compete against the standard MSE loss. For some problems, MSE loss will be optimal and, for the other cases, probably the amplitude shifting loss makes sense. However, how to decide which one to use?
TILDE-Q: A TRANSFORMATION INVARIANT LOSS FUNCTION FOR TIME-SERIES FORECASTING Anonymous authors Paper under double-blind review ABSTRACT Time-series forecasting has gained increasing attention in the field of artificial intelligence due to its potential to address real-world problems across various domains, includin...
Fq8tKtjACC
Another point of contention is the authors' assertion that phi-1 consumed less compute for training. They overlook the computational resources expended in creating their training data, and more importantly, the compute required to train the foundational LLMs.
TEXTBOOKS ARE ALL YOU NEED Anonymous authors Paper under double-blind review ABSTRACT We introduce phi-1, a new large language model for code, with significantly smaller size than competing models: phi-1 is a Transformer-based model with 1.3B parameters, trained for 4 days on 8 A100s, using a selection of “textbook ...
WRxCuhTMB2
The paper has an entire section dedicated to training epistemic injections, but this is never mentioned before as a part of their contribution or utility. The beginning of the paper makes it seem like they offer an evaluation of modern methods, but this along with the meta-model seem to be proposing training procedures...
EVALUATION METHODOLOGY FOR DISENTANGLED UNCERTAINTY QUANTIFICATION ON REGRESSION MODELS Anonymous authors Paper under double-blind review ABSTRACT The lack of an acceptable confidence level associated with the predictions of Machine Learning (ML) models may inhibit their deployment and usage. A practical way to avoi...
SzV37yefM4
If contrastive decoding improves overall generation quality, it should ideally exhibit some improvement in the results without the presence of CoT in the prompts. Do you have any insights on why this is not happening?
CONTRASTIVE DECODING IMPROVES REASONING IN LARGE LANGUAGE MODELS Anonymous authors Paper under double-blind review ABSTRACT We demonstrate that Contrastive Decoding – a simple, computationally light, and training-free text generation method proposed by Li et al 2022 – achieves large out-of-the-box improvements over ...
dRel8fuUK4
While reference records are not always a direct input in other membership inference attacks, is there any way to assess whether the competing methods would depend more strongly or weakly on the number of reference records, compared to the new attack?
LOW-COST HIGH-POWER MEMBERSHIP INFERENCE BY BOOSTING RELATIVITY Anonymous authors Paper under double-blind review ABSTRACT We present a robust membership inference attack (RMIA) that amplifies the distinction between population data and the training data on any target model, by effectively leveraging both reference ...
IJBsKYXaH4
Regarding the measures COV and MAT. If we assume that a molecule has, e.g., 3 major conformers which are all very close in RMSD. Wouldn’t a model that samples always only one conformer achieve a COV of 1, even though it has never generated the other 2? Also, the definition of MAT seems odd, is there sum over S_r and ma...
MOLECULAR CONFORMATION GENERATION VIA SHIFTING SCORES Anonymous authors Paper under double-blind review ABSTRACT Molecular conformation generation, a critical aspect of computational chemistry, involves producing the three-dimensional conformer geometry for a given molecule. Generating molecular conformation via dif...
QLoepRnoue
In the formulation on decoding, (i.e., equation between eq. (2) and eq.(3)), can you please clarify on why orthogonality property ensures that $E_X(x_i) ⊘ E_X(x_0) $ will produce a vector orthogonal to $E_X(x_0)$ when the distance between two samples is large? Also what does the noise mean? Does it mean that it’s near ...
Decodable and Sample Invariant Continuous Object Encoder Dehao Yuan, Furong Huang, Cornelia Fermüller & Yiannis Aloimonos Department of Computer Science University of Maryland College Park, MD 20740, USA {dhyuan, furongh, fermulcm, jyaloimo}@umd.edu Abstract We propose Hyper-Dimensional Function Encoding (HDFE). Giv...
RNgZTA4CTP
No intuition for the 2-step update when the theoretical assumptions are broken:** The paper leans heavily on asymptotic intuitions, but a lot of the wins in 4.3 and 4.4 seem to come from sample efficiency. Is there any intuition for this?
Best Possible Q-Learning Anonymous authors Paper under double-blind review Abstract Fully decentralized learning, where the global information, i.e., the actions of other agents, is inaccessible, is a fundamental challenge in cooperative multi-agent reinforcement learning. However, the convergence and optimality of ...
Zw8YxUWL4R
Overall, I feel the step-wise operation is worth further investigating. To be more specific, my concern is that the statement/the proposed empirical finding “the coarse inner layers of U-Net affecting the shape of the generated image, and the outer layers affecting the style and appearance” may be inaccurate, which is ...
P+: Extended Textual Conditioning in Text-to-Image Generation Anonymous authors Paper under double-blind review Abstract We introduce an Extended Textual Conditioning space in text-to-image diffusion models, referred to as $P_+$. This space consists of multiple textual conditions, derived from per-layer prompts, eac...
m7aPLHwsLr
The low success rates of attacks (especially GAMMA) might be due to a wrong initialisation. In the appendix, it is written that 200 as population size and query are used, but the number of queries for the GAMMA attack are computed as population_size * iterations. Also, the number of used sections is missing (which is a...
DRSM: De-Randomized Smoothing on Malware Classifier Providing Certified Robustness Shoumik Saha, Wenxiao Wang, Yigitcan Kaya, Soheil Feizi & Tudor Dumitras {smksaha, wwx, cankaya, sfeizi, tudor}@umd.edu Department of Computer Science University of Maryland - College Park Abstract Machine Learning (ML) models have be...
dl0u4ODCuW
The reviewer might have misunderstood something, but the problem described in Section 4.1 could easily avoided if the authors used a hash table for implementing the search algorithms. (Using a hash table is a standard technique for proof-number search.) If a hash table is used, a cycle could be easily detected, so it i...
RETRO-FALLBACK: RETROSYNTHETIC PLANNING IN AN UNCERTAIN WORLD Austin Tripp\textsuperscript{1}*, Krzysztof Maziarz\textsuperscript{2}, Sarah Lewis\textsuperscript{2}, Marwin Segler\textsuperscript{2}, José Miguel Hernández-Lobato\textsuperscript{1} \textsuperscript{1}University of Cambridge \textsuperscript{2}Microsof...
RIbH5ekQpr
- The MPL2D metric measures the mean euclidean distance between image and caption embeddings, meant to capture the semantic diversity to justify the aforementioned diversity claims. There may be some concerns regarding this formulation which I would like to give the authors a chance to verify any possible misconception...
IMP: Benchmarking Image Polysemy in Vision-Language Models Anonymous authors Paper under double-blind review Abstract Current vision-language models predominantly use contrastive losses to learn from the co-occurrence of image and text. While effective for certain tasks, this approach assumes semantic equivalence be...
ZS4m74kZpH
If I understand correctly, you use the irrelevant context (e.g., in the single-hop case) to train the LM to answer the question by ignoring the context. Isn't this (almost) the definition of hallucination? The resulting LM will produce information not grounded in any passages. Isn't it better to abstain / request a new...
MAKING RETRIEVAL-AUGMENTED LANGUAGE MODELS ROBUST TO IRRELEVANT CONTEXT Ori Yoran\textsuperscript{1} \quad Tomer Wolfson\textsuperscript{1,2} \quad Ori Ram\textsuperscript{1} \quad Jonathan Berant\textsuperscript{1} \textsuperscript{1}Tel Aviv University, \textsuperscript{2}Allen Institute for AI \{ori.yoran, ori.ram,...
99tKiMVJhY
What is the particular difficulty of solving the Dec-POMFC system, and how does the proposed method solve such difficulty? It would be easier to follow the paper if these questions were explicitly explained.
Learning Decentralized Partially Observable Mean Field Control for Artificial Collective Behavior Kai Cui, Sascha Hauck, Christian Fabian, Heinz Koepl Dept. of Electrical Engineering and Information Technology, Technische Universität Darmstadt {kai.cui, heinz.koepl}@tu-darmstadt.de Abstract Recent reinforcement lear...
JAfGlmRBTU
The introduction raises several points that existing models fail at, e.g., “the parse tree could switch among multiple reasonable forms even given a single scene”, i.e., “correct” parsing is context-dependent. While this is true, the proposed model also does not deal with this (or does it?)
REPRESENTING PART-WHOLE HIERARCHY WITH COORDINATED SYNCHRONY IN NEURAL NETWORKS Anonymous authors Paper under double-blind review ABSTRACT Human vision flexibly extracts part-whole hierarchy from visual scenes. However, how can a neural network with a fixed architecture parse an image into a part-whole hierarchy tha...
GDdxmymrwL
There is some notable performance gap of *Corex* variants. Such as *Corex-Review-Code* v.s. other variants for GSM-Hard in Table 2, and for Repeat Copy in Table 4. Any intuition or explanations on this?
Corex: Pushing the Boundaries of Complex Reasoning through Multi-Model Collaboration Anonymous authors Paper under double-blind review Abstract Large Language Models (LLMs) are evolving at an unprecedented pace and have exhibited considerable capability in the realm of natural language processing (NLP) with world kn...
ijoqFqSC7p
- From looking at the generated videos, although the proposed method can more cleanly generate longer videos, it seems that the spatial structure of the video (e.g. location of a cat) is very similar throughout the entire video. I believe this may be due to the repetitive nature of shuffled noise reptitions which are g...
FREE NOISE: TUNING-FREE LONGER VIDEO DIFFUSION VIA NOISE RESCHEDULING Haonan Qiu1*, Menghan Xia2*, Yong Zhang2, Yingqing He2,3, Xintao Wang2, Ying Shan2, Ziwei Liu1* 1Nanyang Technological University 2Tencent AI Lab 3Hong Kong University of Science and Technology ABSTRACT With the availability of large-scale video ...
zqVvdn0NQM
Using different inputs for the decision tree (DT) model (which uses a three dimensional input consisting of mean (R,G,B) values) and the deep learning (DL) models (which use raw images) does not seem to allow a fair comparison between their explanations.
STOP OVERKILLING SIMPLE TASKS WITH BLACK-BOX MODELS, USE MORE TRANSPARENT MODELS INSTEAD Anonymous authors Paper under double-blind review ABSTRACT The ability of deep learning-based approaches to extract features autonomously from raw data while outperforming traditional methods has led to several breakthroughs in ...
AJBkfwXh3u
Additionally, the computational intensity of introducing temporal masks, which could be exacerbated by the incorporation of contrastive learning (VGAE is known to be computationally demanding), is not addressed.
Causality-Inspired Spatial-Temporal Explanations for Dynamic Graph Neural Networks Kesen Zhao City University of Hong Kong Hong Kong, China kesenzhao2-c@my.cityu.edu.hk Liang Zhang * Shenzhen Research Institute of Big Data Guangdong, China zhangliang@sribd.cn Abstract Dynamic Graph Neural Networks (DyGN...
uHVIxJGwr4
Could the authors clarify what constitutes a 'transition' in this context? Does the transition include (s,a,s’) even when FSB is not employed in VHB, which is 0.05 times? Do you discard any transition? How is it ensured that you explore a wide array of instances before 100K transitions are collected?
Learning to Branch with Offline Reinforcement Learning Anonymous authors Paper under double-blind review Abstract Mixed Integer Linear Program (MILP) solvers are mostly built upon a branch-and-bound (B&B) algorithm, where the efficiency of traditional solvers heavily depends on hand-crafted heuristics for branching....
ZA9XUTseA9
similarly, in the experiment, why is the perturbed 1 norm close to 0 at convergence? It seems the authors are performing early stopping, but that precisely means that implicit regularization is not happening, and that the model overfits.
On the Implicit Bias of Adam Anonymous authors Paper under double-blind review Abstract In previous literature, backward error analysis was used to find ordinary differential equations (ODEs) approximating the gradient descent trajectory. It was found that finite step sizes implicitly regularize solutions because te...
x1ptaXpOYa
Could you provide a more detailed breakdown of the types and sources of documents included in the ADoPD dataset? Understanding the diversity in terms of document genres, geographical origins, and linguistic variations would offer more insight into its applicability and robustness.
ADOPD: A LARGE-SCALE DOCUMENT PAGE DECOMPOSITION DATASET Jiuxiang Gu\textsuperscript{1}\textsuperscript{*} Xiangxi Shi\textsuperscript{2} Jason Kuen\textsuperscript{1} Lu Qi\textsuperscript{3} Ruiyi Zhang\textsuperscript{1} Anqi Liu\textsuperscript{4} Ani Nenkova\textsuperscript{1} Tong Sun\textsuperscript{1} \textsup...
X1lDOv09hG
The paper only considers the linear score estimator and derives the optimal closed-form solution. How do you know the ground truth score function is linear? For a general score function, can we still have high variance parameters?
HIGH VARIANCE SCORE FUNCTION ESTIMATES HELP DIFFUSION MODELS GENERALIZE Anonymous authors Paper under double-blind review ABSTRACT How do diffusion-based generative models generalize beyond their training set? In particular, do they perform something similar to kernel density estimation? If so, what is the kernel, a...
nW0sCc3LLN
Similarly, other works on generative MI have previously shown the opposite results to what you share here [3] - when such mitigation (in that case split learning) is present, it is actually beneficial for the attacker to use a smaller section of the model (due to the ease of reconstruction).
Model Inversion Robustness: Can Transfer Learning Help? Anonymous authors Paper under double-blind review Abstract Model Inversion (MI) attacks aim to reconstruct private training data by abusing access to machine learning models. Contemporary MI attacks have achieved impressive attack performance posing serious thr...
bgyWXX8HCk
How does this work compare: Efficient Representation of Numerical Optimization Problems for {SNARKs}. Angel et al. 2022. I also see several papers in federated learning that leverage SNARKs for enhancing trust. If models involving softmax or similar steps to the presently-experimented ones were used, I imagine they wou...
TRUSTLESS AUDITS WITHOUT REVEALING DATA OR MODELS Anonymous authors Paper under double-blind review ABSTRACT There is an increasing conflict between business incentives to hide models and data as trade secrets, and the societal need for algorithmic transparency. For example, a rightsholder wishing to know whether th...
ePOjNlOjLC
According to Table 1, the proposed method is inferior to SD inpainting on both performance and efficiency. The only superiority of the proposed method is training free. However, since it needs cyclical diffusion & denoising, its inference cost is higher than SD inpainting. The superiority may be weakened.
Diffusion in Diffusion: Cyclic One-Way Diffusion for Text-Vision-Conditioned Generation Ruoyu Wang\textsuperscript{1}\textsuperscript{*}, Yongqi Yang\textsuperscript{1}\textsuperscript{*}, Zhihao Qian\textsuperscript{1}, Ye Zhu\textsuperscript{2}, Yu Wu\textsuperscript{1}\textsuperscript{†} \textsuperscript{1} School ...
2DbVeuoa6a
The operators $T$ and $T^{-1}$ are not detailed. What is their complexity with respect to the input length ? How do you choose the collocation points ? Does the number depend on the difficulty level of the PDE ?
Neural Spectral Methods: Self-supervised learning in the spectral domain Yiheng Du, Nithin Chalapathi, Aditi S. Krishnapriyan {yihengdu, nithinc, aditik1}@berkeley.edu University of California, Berkeley Abstract We present Neural Spectral Methods, a technique to solve parametric Partial Differential Equations (PDEs)...
sTf7mXhTVt
To minimize $F(x_0 + \delta, y)$ under some constraints, the authors solve the problem (5) derived from inequalities originating from the smoothness of $F$ and the Lipschitz continuity of the gradient. However, it is important to note that problem (5) does not necessarily entail the minimization of $F(x_0 + \delta, y)$...
Query Efficient Black-Box Adversarial Attack with Automatic Region Selection Anonymous authors Paper under double-blind review Abstract Deep neural networks (DNNs) have been shown to be vulnerable to black-box attacks in which small perturbations are added to input images without accessing any internal information o...
KqTzfiNjWU
If I understand well, DPS is adapted to the proposed framework. But does this make sense? In particular, while I understand that DPS is not well suited for dehazing as well as deraining, it performs rather well on deblurring - but Figure 7 shows no difference between input images and output images for DPS. Could the au...
RESTORER GUIDED DIFFUSION MODELS FOR VARIATIONAL INVERSE PROBLEMS Anonymous authors Paper under double-blind review ABSTRACT Diffusion models have made remarkable progress in solving various inverse problems, attributing to the generative modeling capability of the data manifold. Posterior sampling from the conditio...
x5LvBK43wg
How does the graph construction technique manages the class imbalance that might be present in the unlabeled target data? Related to the discussion in section 3.2 about initialization of prototypes and constructing a prototype graph.
PROGRAM: PROtotype GRAph Model based Pseudo-Label Learning for Test-Time Adaptation Haopeng Sun1*, Lumin Xu3 Sheng Jin4,2 Ping Luo4,5 Chen Qian1,2 Wentao Liu2 1 Department of Computer Science and Technology, Tsinghua University, Beijing, China 2 SenseTime Research and Tetras.AI 3 The Chinese University of Hong Kong 4...
WM5G2NWSYC
This work often leverages metalearning as a driving motivator. But how crucial is a Meta-Learning approach in light of works such as [7,8] which highlight limited importance of a full meta-learning objective for few-shot transfer?
Projected Subnetworks Scale Adaptation Anonymous authors Paper under double-blind review Abstract Large models support great zero-shot and few-shot capabilities. However, updating these models on new tasks can break performance on previous seen tasks and their zero/few-shot unseen tasks. Our work explores how to upd...
2fSyBPBfBs
Since it is possible to encounter the bad $x_{out}$ that {$||s||: s \in \partial_{\delta} \varphi(x_{out})$} is large, can you explain how to choose the output in Algorithm 2 in detail in the experiment?
Bilevel Optimization without Lower-Level Strong Convexity from the Hyper-Objective Perspective Anonymous authors Paper under double-blind review Abstract Bilevel optimization reveals the inner structure of otherwise oblique optimization problems, such as hyperparameter tuning, neural architecture search, and meta-le...
sOXKeeVxqW
However, the paper falls short in providing a comprehensive explanation regarding the rationale behind their selection and the potential implications if alternative encoders were to be used. Elaborating on the specific reasons for choosing these models and discussing the potential consequences of substituting them with...
MOleSG: A Multi-Modality Molecular Pre-training Framework by Joint Non-overlapping Masked Reconstruction of SMILES and Graph Anonymous authors Paper under double-blind review Abstract Self-supervised pre-training plays an important role in molecular representation learning because labeled molecular data are usually ...
P895PSh41Z
* Although it is repeatedly mentioned that one of the primary motivations for avoiding a model-based algorithm is that they struggle with stochastic environments, all the experiments are conducted on deterministic tasks.
Relaxed State-Adversarial Offline Reinforcement Learning: A Leap Towards Robust Model-Free Policies from Historical Data Anonymous authors Paper under double-blind review Abstract Offline reinforcement learning (RL) targets the development of top-tier policies from historical data, eliminating the need for environme...
rUH2EDpToF
As far as I understood, there's no guarantee that the model satisfies the marginalization self-consistency constraint (Eq. 5), and therefore the model compares to any other dealing with approximate marginal inference, such as [1] [2].
GENERATIVE MARGINALIZATION MODELS Anonymous authors Paper under double-blind review ABSTRACT We introduce marginalization models (MAMs), a new family of generative models for high-dimensional discrete data. They offer scalable and flexible generative modeling with tractable likelihoods by explicitly modeling all ind...
wXpSidPpc5
Not much details is provided in the main text regarding how we train such a beast. I must say this looks quite daunting to me how I would train a NODE along my transformer model. I guess it would help to have some explanations to it.
CLEX: Continuous Length Extrapolation for Large Language Models Guanzheng Chen\textsuperscript{1,2,3,*} Xin Li\textsuperscript{2,3,†} Zaiqiao Meng\textsuperscript{4} Shangsong Liang\textsuperscript{1,5,†} Lidong Bing\textsuperscript{2,3} \textsuperscript{1}Sun Yat-sen University \textsuperscript{2}DAMO Academy, Alibab...
o2IEmeLL9r
I understand the motivation behind the KL weight and that neither alpha=0 (no prior) nor too large of a weight are desirable. However, it appears that the authors choose to train the high-level policy from scratch and only leverage the goal prior to guide exploration. Given that the goal prior and high-level policy sha...
Pre-Training Goal-based Models for Sample-Efficient Reinforcement Learning Haoqi Yuan\textsuperscript{1}, Zhancun Mu\textsuperscript{2}, Feiyang Xie\textsuperscript{2}, Zongqing Lu\textsuperscript{1,3†} \textsuperscript{1}School of Computer Science, Peking University \textsuperscript{2}Yuanpei College, Peking Univers...
9nXgWT12tb
I think the most fair comparison is transformer vs transformer +CBA where the transformer has the same number of heads as the transformer +CBA (when we count both temporal attention and correlated attention heads). Does the transformer in Table 2 has exact same head as the transformer +CBA?
CORRELATED ATTENTION IN TRANSFORMERS FOR MULTIVARIATE TIME SERIES Anonymous authors Paper under double-blind review ABSTRACT Multivariate time series (MTS) analysis prevails in real-world applications such as finance, climate science and healthcare. The various self-attention mechanisms, the backbone of the state-of...
13D1zn0mpd
In the proposed method, several points of clarification regarding the comparison with LoRA emerge. Firstly, it would be beneficial to understand what distinguishes the proposed method from LoRA. While the primary technique appears to focus on reducing the rank of $A_t B_t^{\top}$. Secondly, when referencing Table 4, on...
EFFECTIVE AND PARAMETER-EFFICIENT REUSING FINE-TUNED MODELS Anonymous authors Paper under double-blind review ABSTRACT Many pre-trained large-scale models provided online have become highly effective in transferring to downstream tasks. At the same time, various task-specific models fine-tuned on these pre-trained m...
ufvwhR3XmN
The tradeoff study between temporal context and spectral context is not able to lead such conclusion that higher frequency domain resolution provideds more benefits compared higher time domain resolution, as the results of these two setting are very close in the test set (20.80 vs. 20.66).
A JOINT SPECTRO-TEMPORAL RELATIONAL THINKING BASED ACOUSTIC MODELING FRAMEWORK Anonymous authors Paper under double-blind review ABSTRACT Relational thinking refers to the inherent ability of humans to form mental impressions about relations between sensory signals and prior knowledge, and subsequently incorporate t...
IhD1rBHhDy
The paper assumes that chemical structures with similar functionality should cluster close to each other in the structural space based on molecular fingerprints; however, this doesn’t necessarily have to be the case - you can have stereoisomers with different functional properties, and you can have chemical compounds w...
MINING PATENTS WITH LARGE LANGUAGE MODELS ELUCIDATES THE CHEMICAL FUNCTION LANDSCAPE Anonymous authors Paper under double-blind review ABSTRACT The fundamental goal of small molecule discovery is to generate chemicals with target functionality. While this often proceeds through structure-based methods, we set out to...
K7l94Z81bH
Given the heterogeneity of agents in your setup, how does RLD3 ensure fair allocation of rewards and prevent potential domination by certain groups or agents, which could lead to suboptimal overall system performance?
Sparsity-Aware Grouped Reinforcement Learning for Designated Driver Dispatch Anonymous authors Paper under double-blind review Abstract Designated driving service is a fast-growing market that provides drivers to transport customers in their own cars. The main technical challenge in this business is the design of dr...
uvFhCUPjtI
The timespan of edges is a natural attribute of a temporal graph. Some recurrent works [1] [2] show that embedding the timespan of edges is important, especially in the sequential recommendation. I am wondering whether EFT could embed the timespan of edges and how. Such discussion may help the audience have better idea...
Beyond Spatio-Temporal Representations: Evolving Fourier Transform for Temporal Graphs Anson Bastos\textsuperscript{1,2}, Kuldeep Singh\textsuperscript{4}, Abhishek Nadgeri\textsuperscript{3}, Manish Singh\textsuperscript{2}, Toyotaro Suzumura\textsuperscript{5} \textsuperscript{1}HERE Technologies, India \textsupersc...
vzvCaYFTLq
Currently, only V100 is considered as the target device. However, newer generations of GPUs are rapidly developing and providing more effective support for lower-bit inference and memory consumption (e.g., H100/A100).
SAPLING: SUCCESSIVE ADAPTATION AND COMPRESSION WITH LAYER DROPPING FOR LLMs Anonymous authors Paper under double-blind review ABSTRACT Specializing Large language models (LLMs) for local deployment and domain-specific use can deliver state-of-the-art performance while meeting latency and privacy requirements. Howeve...
MCUvAc1GTg
In Section 4.3 in your data augmentation method you propose to permute the node labels of the perturbed graphs. GNNs should either be permutation equivariant or invariant to node permutation, in either case the permutation of node labels in your perturbed graphs should be inconsequential. Could you therefore please mot...
Network Alignment with Transferable Graph Autoencoders Anonymous authors Paper under double-blind review Abstract Network alignment is the task of establishing one-to-one correspondences between the nodes of different graphs and finds a plethora of applications in high-impact domains. However, this task is known to ...
UPvufoBAIs
As mentioned in the article, an observation is that global information is noisy, but some local details are robust. I hope there is rigorous explanation and quantitative analysis here to support this hypothesis.
Source-Free and Image-Only Unsupervised Domain Adaptation for Category Level Object Pose Estimation Prakhar Kaushik Aayush Mishra Adam Kortylewski† Alan Yuille Johns Hopkins University †University of Freiburg and Max-Planck-Institute for Informatics {pkaushil,amishr24,ayuille1}@jh.edu †akortyle@mpi-inf.mpg.de Abst...
uZfjFyPAvn
If authors think Shearlets may have any potential here, providing a discussion might be useful. Most of the approximation error for the pictures in Figure 6 appear to be at locations where the colors change in a small neighborhood. Could a shear matrix be potentially helpful in reducing the error because of its ability...
Implicit Neural Representations and the Algebra of Complex Wavelets T. Mitchell Roddenberry, Vishwanath Saragadam∗, Maarten V. de Hoop, Richard G. Baraniuk Rice University Houston, TX, USA {mitch,mvd2,richb}@rice.edu, vishwanath.saragadam@ucr.edu Abstract Implicit neural representations (INRs) have arisen as useful ...
wPhbtwlCDa
would be great to provide (even 1-sentence) intuition for EPIC (explain the various normalization terms that EPIC paper explains), because it is crucial for understanding your definition of canonicalization function
STARC: A GENERAL FRAMEWORK FOR QUANTIFYING DIFFERENCES BETWEEN REWARD FUNCTIONS Joar Skalse Department of Computer Science Future of Humanity Institute Oxford University joar.skalse@cs.ox.ac.uk Lucy Farnik University of Bristol Bristol AI Safety Centre lucy.farnik@bristol.ac.uk Sumeet Ramesh Motwani ...
LAEd3kHao9
The proposed languageinformed distributions (LID) can effectively avoid the issue of intra-class variety. However, the authors would better also intepret how to solve the issue of inter-class correlation.
Promoting Language-Informed Distribution for Compositional Zero-Shot Learning Anonymous authors Paper under double-blind review Abstract Compositional zero-shot learning (CZSL) task aims to recognize unseen compositional visual concepts, e.g., sliced tomatoes, where the model is learned only from the seen compositio...
S7j1sNVIm9
The theoretical analysis to the heterogeneity is not convincing. $\sigma_f^2$ is used as a measure of client heterogeneity in the paper, however, it is just an upper-bound (Proposition 1) of some more classical measure of heterogeneity, which means the proposed measure is weaker. In fact, if $l^*$ is chosen to be 0 (as...
Locally Adaptive Federated Learning Anonymous authors Paper under double-blind review Abstract Federated learning is a paradigm of distributed machine learning in which multiple clients coordinate with a central server to learn a model, without sharing their own training data. Standard federated optimization methods...
NqQjoncEDR
Similarly, in Figure 2, for ''Same domain'', ''Diff. class'' and ''Diff. domain + Diff. class'', selective sampling is much better than selective Mixup, does this indicate that vanilla Mixup is harmful in this case? Such observations are more evident in Figure 8. The authors have not discussed the reasons behind the oc...
SELECTIVE MIXUP HELPS WITH DISTRIBUTION SHIFTS, BUT NOT (ONLY) BECAUSE OF MIXUP Anonymous authors Paper under double-blind review ABSTRACT Context. Mixup is a highly successful technique to improve generalization of neural networks by augmenting the training data with combinations of random pairs. Selective mixup is...
QIrYb3Vlze
It seems like the $H$ space in [1] (considered a reliable semantic space by the authors) is obtained from the pre-trained Diffusion Models. If Diffusion Models are trained from scratch with an additional objective, how do the authors ensure that the $H$ space in [1] and the $H$ space in this paper have similar properti...
ISOMETRIC REPRESENTATION LEARNING FOR DISENTANGLING LATENT SPACE OF DIFFUSION MODELS Anonymous authors Paper under double-blind review ABSTRACT Diffusion models have made remarkable progress in capturing and reproducing real-world data. Despite their success and further potential, their latent space, the core of dif...
vQqJJzL2Jf
Hence, the difference in the PDE behaviour may also come from how long a certain trajectory is followed, not intrinsically from what kind of equation it is. Then, a user of the method may feel safe to extrapolate a given PDF that has seemed to be
Understanding and Mitigating Extrapolation Failures in Physics-Informed Neural Networks Anonymous authors Paper under double-blind review Abstract Physics-informed Neural Networks (PINNs) have recently gained popularity due to their effective approximation of partial differential equations (PDEs) using deep neural n...
krIOxfqsOh
What is the architecture of MaskMA? Is it also based on a Decision Transformer? I didn't find value as input in the transformer architecture shown in Figure 2. How did you train MaskMA? Is it similar to supervised learning in a Decision transformer?
Masked Pretraining for Multi-Agent Decision Making Anonymous authors Paper under double-blind review Abstract Building a single generalist agent with zero-shot capability has recently sparked significant advancements in decision-making. However, extending this capability to multi-agent scenarios presents challenges....
4uaogMQgNL
Quantitatively, the UpFusion 3D model has much better numbers than the 2D model, but visually it loses a lot of geometric details compared to the 2D results. Is it limited by the representation power of the 3D NeRF? Or is it because the learned features are not very view-consistent?
UpFusion: Novel View Diffusion from Unposed Sparse View Observations Anonymous authors Paper under double-blind review Figure 1: 3D Inference from Unposed Sparse views. Given a sparse set of input images without associated camera poses, our proposed system UpFusion allows recovering a 3D representation and synthesizi...
9Gvs64deOj
Can the authors explain why it is particularly interesting to train clients in the wireless setting and include the noisy channel in their estimation rather than using wireless communication protocols to encode and decode messages (if needed) and then conducting FedAvg or variants of FedAvg? By construction of the algo...
Rendering Wireless Environments Useful for Gradient Estimators: A Zero-Order Stochastic Federated Learning Method Anonymous authors Paper under double-blind review Abstract Federated learning (FL) is a novel approach to machine learning that allows multiple edge devices to collaboratively train a model without discl...
qcigbR1UYA
Did the authors consider extending the work to more general tests where a test consists of passing $Y$ through a noisy channel of input alphabet $\mathcal{Y}$ (the set of possible value of $Y$) and of output alphabet \{0,1\}?
Performance Bounds for Active Binary Testing with Information Maximization Anonymous authors Paper under double-blind review Abstract In many applications like experimental design, group testing, medical diagnosis, and active testing, the state of a random variable $Y$ is revealed by successively observing the outco...
w8eCnnq57m
**Computational cost of optimization**: Unlike In-Context Learning, `LoraHub` does not need to process additional tokens hence a reduced inference cost. However, it also adds the cost to optimize the combination weights $w$ on the input few-shot samples, in particular when many upstream tasks are available. It would be...
LORAHub: Efficient Cross-Task Generalization via Dynamic LoRA Composition Anonymous authors Paper under double-blind review Abstract Low-rank adaptations (LoRA) are often employed to fine-tune large language models (LLMs) for new tasks. This paper investigates LoRA composability for cross-task generalization and int...
jU3zRzUBiD
Previous work, such as CryptoNAS and Sphynx, has already explored this concept by maintaining a constant ReLU count per layer, which in turn increases the FLOPs count (however, these studies did not provide detailed FLOPs count information)
COMPENSATING FOR NONLINEAR REDUCTION WITH LINEAR COMPUTATIONS IN PRIVATE INFERENCE Anonymous authors Paper under double-blind review ABSTRACT Increasingly serious data privacy concerns and strict regulations have recently posed significant challenges to machine learning, a field that hinges on high-performance proce...
yBZd6mCWXd
However, the authors omit comparing it with the version that removes alpha in the ablation study (refer to Table 7). This omission makes it challenging to discern the contribution of this modification.
WI3D: Weakly Incremental 3D Detection via Visual Prompts Anonymous authors Paper under double-blind review Abstract Class-incremental 3D object detection demands a 3D detector to locate and recognize novel categories in a stream fashion, while not forgetting its previously learned knowledge. However, existing method...
NSVtmmzeRB
Are there any architectural / data preprocessing differences between the diffusion models (e.g. EDM) and the proposed GeoBFN? Are all improvements in performance attributable to the new training / sampling algorithm given by the Bayesian Flow Networks formulation [1]? Though the derivation is different, the training lo...
Unified Generative Modeling of 3D Molecules via Bayesian Flow Networks Yuxuan Song\textsuperscript{1}\textsuperscript{*}, Jingjing Gong\textsuperscript{1}\textsuperscript{*}, Hao Zhou\textsuperscript{1}, Mingyue Zheng\textsuperscript{2}, Jingjing Liu\textsuperscript{1} & Wei-Ying Ma\textsuperscript{1} \textsuperscript...
Free AI Image Generator No sign-up. Instant results. Open Now