Getting ArXiv Sanity Papers Via RSS

[
  {
    "authors": "Takeshi Inagaki",
    "id": "2205.11509",
    "summary": "In natural language processing (NLP), labeling on regions of text, such as\nwords, sentences and paragraphs, is a basic task. In this paper, label is\ndefined as map between mention of entity in a region on text and context of\nentity in a broader region on text containing the mention. This definition\nnaturally introduces linkage of entities induced from inclusion relation of\nregions, and connected entities form a graph representing information flow\ndefined by map. It also enables calculation of information loss through map\nusing entropy, and entropy lost is regarded as distance between two entities\nover a path on graph.",
    "tags": "cs.CL, cs.AI, cs.LG",
    "thumb_url": "static/thumb/2205.11509.jpg",
    "time": "May 23 2022",
    "title": "Information Propagation by Composited Labels in Natural Language\n  Processing",
    "utags": [],
    "weight": 0.8466816017059265
  },
  {
    "authors": "Randall Balestriero, Yann LeCun",
    "id": "2205.11508",
    "summary": "Self-Supervised Learning (SSL) surmises that inputs and pairwise positive\nrelationships are enough to learn meaningful representations. Although SSL has\nrecently reached a milestone: outperforming supervised methods in many\nmodalities... the theoretical foundations are limited, method-specific, and\nfail to provide principled design guidelines to practitioners. In this paper,\nwe propose a unifying framework under the helm of spectral manifold learning to\naddress those limitations. Through the course of this study, we will rigorously\ndemonstrate that VICReg, SimCLR, BarlowTwins et al. correspond to eponymous\nspectral methods such as Laplacian Eigenmaps, Multidimensional Scaling et al.\nThis unification will then allow us to obtain (i) the closed-form optimal\nrepresentation for each method, (ii) the closed-form optimal network parameters\nin the linear regime for each method, (iii) the impact of the pairwise\nrelations used during training on each of those quantities and on downstream\ntask performances, and most importantly, (iv) the first theoretical bridge\nbetween contrastive and non-contrastive methods towards global and local\nspectral embedding methods respectively, hinting at the benefits and\nlimitations of each. For example, (a) if the pairwise relation is aligned with\nthe downstream task, any SSL method can be employed successfully and will\nrecover the supervised method, but in the low data regime, SimCLR or VICReg\nwith high invariance hyper-parameter should be preferred; (b) if the pairwise\nrelation is misaligned with the downstream task, BarlowTwins or VICReg with\nsmall invariance hyper-parameter should be preferred.",
    "tags": "cs.LG, cs.AI, cs.CV, math.SP, stat.ML",
    "thumb_url": "static/thumb/2205.11508.jpg",
    "time": "May 23 2022",
    "title": "Contrastive and Non-Contrastive Self-Supervised Learning Recover Global\n  and Local Spectral Embedding Methods",
    "utags": [],
    "weight": 1.0686954905948152
  },
  {
    "authors": "Dongruo Zhou, Quanquan Gu",
    "id": "2205.11507",
    "summary": "Recent studies have shown that episodic reinforcement learning (RL) is not\nmore difficult than contextual bandits, even with a long planning horizon and\nunknown state transitions. However, these results are limited to either tabular\nMarkov decision processes (MDPs) or computationally inefficient algorithms for\nlinear mixture MDPs. In this paper, we propose the first computationally\nefficient horizon-free algorithm for linear mixture MDPs, which achieves the\noptimal $\\tilde O(d\\sqrt{K} +d^2)$ regret up to logarithmic factors. Our\nalgorithm adapts a weighted least square estimator for the unknown transitional\ndynamic, where the weight is both \\emph{variance-aware} and\n\\emph{uncertainty-aware}. When applying our weighted least square estimator to\nheterogeneous linear bandits, we can obtain an $\\tilde O(d\\sqrt{\\sum_{k=1}^K\n\\sigma_k^2} +d)$ regret in the first $K$ rounds, where $d$ is the dimension of\nthe context and $\\sigma_k^2$ is the variance of the reward in the $k$-th round.\nThis also improves upon the best-known algorithms in this setting when\n$\\sigma_k^2$\u0027s are known.",
    "tags": "cs.LG, math.OC, stat.ML",
    "thumb_url": "static/thumb/2205.11507.jpg",
    "time": "May 23 2022",
    "title": "Computationally Efficient Horizon-Free Reinforcement Learning for Linear\n  Mixture MDPs",
    "utags": [],
    "weight": 1.0688575276318522
  }
]

#!/bin/bash
if curl "https://validator.w3.org/feed/check.cgi?url=${FEED_URL}" | grep "This is a valid RSS feed."
then
    echo "The feed is valid."
    exit 0
else
    echo "The feed is invalid."
    exit 1
fi