Picture for Yusuke Iwasawa

Yusuke Iwasawa

Rethinking Evaluation of Sparse Autoencoders through the Representation of Polysemous Words

Add code
Jan 09, 2025
Figure 1 for Rethinking Evaluation of Sparse Autoencoders through the Representation of Polysemous Words
Figure 2 for Rethinking Evaluation of Sparse Autoencoders through the Representation of Polysemous Words
Figure 3 for Rethinking Evaluation of Sparse Autoencoders through the Representation of Polysemous Words
Figure 4 for Rethinking Evaluation of Sparse Autoencoders through the Representation of Polysemous Words
Viaarxiv icon

ADOPT: Modified Adam Can Converge with Any $β_2$ with the Optimal Rate

Add code
Nov 05, 2024
Figure 1 for ADOPT: Modified Adam Can Converge with Any $β_2$ with the Optimal Rate
Figure 2 for ADOPT: Modified Adam Can Converge with Any $β_2$ with the Optimal Rate
Figure 3 for ADOPT: Modified Adam Can Converge with Any $β_2$ with the Optimal Rate
Figure 4 for ADOPT: Modified Adam Can Converge with Any $β_2$ with the Optimal Rate
Viaarxiv icon

Which Programming Language and What Features at Pre-training Stage Affect Downstream Logical Inference Performance?

Add code
Oct 09, 2024
Figure 1 for Which Programming Language and What Features at Pre-training Stage Affect Downstream Logical Inference Performance?
Figure 2 for Which Programming Language and What Features at Pre-training Stage Affect Downstream Logical Inference Performance?
Figure 3 for Which Programming Language and What Features at Pre-training Stage Affect Downstream Logical Inference Performance?
Figure 4 for Which Programming Language and What Features at Pre-training Stage Affect Downstream Logical Inference Performance?
Viaarxiv icon

Answer When Needed, Forget When Not: Language Models Pretend to Forget via In-Context Knowledge Unlearning

Add code
Oct 01, 2024
Figure 1 for Answer When Needed, Forget When Not: Language Models Pretend to Forget via In-Context Knowledge Unlearning
Figure 2 for Answer When Needed, Forget When Not: Language Models Pretend to Forget via In-Context Knowledge Unlearning
Figure 3 for Answer When Needed, Forget When Not: Language Models Pretend to Forget via In-Context Knowledge Unlearning
Figure 4 for Answer When Needed, Forget When Not: Language Models Pretend to Forget via In-Context Knowledge Unlearning
Viaarxiv icon

Language Models Do Hard Arithmetic Tasks Easily and Hardly Do Easy Arithmetic Tasks

Add code
Jun 04, 2024
Figure 1 for Language Models Do Hard Arithmetic Tasks Easily and Hardly Do Easy Arithmetic Tasks
Figure 2 for Language Models Do Hard Arithmetic Tasks Easily and Hardly Do Easy Arithmetic Tasks
Figure 3 for Language Models Do Hard Arithmetic Tasks Easily and Hardly Do Easy Arithmetic Tasks
Figure 4 for Language Models Do Hard Arithmetic Tasks Easily and Hardly Do Easy Arithmetic Tasks
Viaarxiv icon

On the Multilingual Ability of Decoder-based Pre-trained Language Models: Finding and Controlling Language-Specific Neurons

Add code
Apr 03, 2024
Figure 1 for On the Multilingual Ability of Decoder-based Pre-trained Language Models: Finding and Controlling Language-Specific Neurons
Figure 2 for On the Multilingual Ability of Decoder-based Pre-trained Language Models: Finding and Controlling Language-Specific Neurons
Figure 3 for On the Multilingual Ability of Decoder-based Pre-trained Language Models: Finding and Controlling Language-Specific Neurons
Figure 4 for On the Multilingual Ability of Decoder-based Pre-trained Language Models: Finding and Controlling Language-Specific Neurons
Viaarxiv icon

Interpreting Grokked Transformers in Complex Modular Arithmetic

Add code
Feb 27, 2024
Viaarxiv icon

Unnatural Error Correction: GPT-4 Can Almost Perfectly Handle Unnatural Scrambled Text

Add code
Nov 30, 2023
Figure 1 for Unnatural Error Correction: GPT-4 Can Almost Perfectly Handle Unnatural Scrambled Text
Figure 2 for Unnatural Error Correction: GPT-4 Can Almost Perfectly Handle Unnatural Scrambled Text
Figure 3 for Unnatural Error Correction: GPT-4 Can Almost Perfectly Handle Unnatural Scrambled Text
Figure 4 for Unnatural Error Correction: GPT-4 Can Almost Perfectly Handle Unnatural Scrambled Text
Viaarxiv icon

Grokking Tickets: Lottery Tickets Accelerate Grokking

Add code
Oct 30, 2023
Figure 1 for Grokking Tickets: Lottery Tickets Accelerate Grokking
Figure 2 for Grokking Tickets: Lottery Tickets Accelerate Grokking
Figure 3 for Grokking Tickets: Lottery Tickets Accelerate Grokking
Figure 4 for Grokking Tickets: Lottery Tickets Accelerate Grokking
Viaarxiv icon

Open X-Embodiment: Robotic Learning Datasets and RT-X Models

Add code
Oct 17, 2023
Figure 1 for Open X-Embodiment: Robotic Learning Datasets and RT-X Models
Figure 2 for Open X-Embodiment: Robotic Learning Datasets and RT-X Models
Figure 3 for Open X-Embodiment: Robotic Learning Datasets and RT-X Models
Figure 4 for Open X-Embodiment: Robotic Learning Datasets and RT-X Models
Viaarxiv icon