and s

By O. Castro-Orgaz, W. Hager. 2019

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

TruthfulQA: Measuring How Models Mimic Human Falsehoods (2021, background): “GPT-Neo/J (Black et al., 2021; Wang and Komatsuzaki, 2021) is a variant of GPT-3 with a different training set (Gao et al., 2020).”
Evaluating the Social Impact of Generative AI Systems in Systems and Society (2023, neutral): “Across modalities, biases can be evaluated using intrinsic and extrinsic methods [91], where the former seeks to evaluate biases within model weights and the latter evaluates the expression of biases in the outputs for downstream tasks (e.”
The Capacity for Moral Self-Correction in Large Language Models (2023, background): “5B parameters for GPT-2 and 11B parameters for T5), and also propose an algorithm for self-debiasing, which has subsequently been shown to be one of the more promising of a variety of debiasing methods [36].”
Semantic Image Synthesis With Spatially-Adaptive Normalization (2019, methodology): “The GANs consist of a generator and a discriminator where the goal of the generator is to produce realistic images so that the discriminator cannot tell the synthesized images apart from real ones.”
Inner Monologue: Embodied Reasoning through Planning with Language Models (2022, background): “One of the remarkable observations in recent machine learning research is that large language models (LLMs) can not only generate fluent textual descriptions, but also appear to have rich internalized knowledge about the world [3, 4, 5, 6, 7].”
Pre-Trained Language Models for Interactive Decision-Making (2022, background): “Several recent papers [36, 17, 15] propose to use LM for policy learning.”
Differentiable Causal Discovery from Interventional Data (2020, background): “As a concrete example, consider the case of gene knockouts/knockdowns experiments in the field of genomics, which aim to understand how specific genes and diseases interact [55].”
Byzantine-Resilient Secure Federated Learning (2020, background): “A related line of work is model poisoning attacks, which are studied in [31], [32].”
Federated Learning with Autotuned Communication-Efficient Secure Aggregation (2019, methodology): “[1] show that even aggressive quantization benefits greatly from pre-processing with a random rotation: for n users, when the rotated update vector x is quantized to a single bit per dimension, a mean squared error”
A Causal Bayesian Networks Viewpoint on Fairness (2018, background): “This effort has led to the emergence of a rich set of fairness definitions [13,16,21,24,38] providing researchers and practitioners with criteria to evaluate existing systems or to design new ones.”
Efficient Structured Pruning and Architecture Searching for Group Convolution (2018, background): “, L1, L2 norms [27, 5, 20, 26], firstorder [29, 3] or second-order [19, 7, 2] gradients, average percentage of zero [26, 11], singular values [32, 28].”
RUDDER: Return Decomposition for Delayed Rewards (2018, methodology): “Since sensitivity analysis substantially hinders learning, we use contribution analysis for backward 62 analysis like contribution-propagation [25], contribution approach [38], excitation backprop [68], 63 layer-wise relevance propagation (LRP) [2], Taylor decomposition [2, 34], or integrated gradie”
Set2Graph: Learning Graphs From Sets (2020, methodology): “Restricting models to be invariant or equivariant to these transformation was shown to be an excellent approach for reducing the number of parameters of models while improving generalization [50, 28, 16, 10, 39, 47, 18, 23, 22, 3, 5, 8, 46, 4, 9, 43, 45, 43].”
High-resolution de novo structure prediction from primary sequence (2022, methodology): “We use the token-dropout scheme as in the ESM-1b model ( 12 ) during pre-training and AdamW ( 45 ) algorithm as our optimizer for OmegaPLM.”
Class-Incremental Continual Learning Into the eXtended DER-Verse (2022, background): “what concerns the main subject of the image): considering the true class, does the model ascribe its score to the expected spatial location? How much does its explanation overlap with the foreground region? To answer these questions, we take into consideration the evaluation protocol proposed in [70”