MétaCan
Menu
Back to cohort

Transfer Learning with Input Reconstruction Loss

2022· article· en· W4315630116 on OpenAlex
Wei Yu

Why this work is in the frame

A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.

affAt least one author lists a Canadian institution in the pinned OpenAlex snapshot.

Bibliographic record

VenueGLOBECOM 2022 - 2022 IEEE Global Communications Conference · 2022
Typearticle
Languageen
FieldComputer Science
TopicDomain Adaptation and Few-Shot Learning
Canadian institutionsUniversity of Toronto
Fundersnot available
KeywordsComputer scienceTransfer of learningArtificial intelligenceArtificial neural networkMachine learningExploitWirelessDeep learningFeature (linguistics)Task (project management)Wireless networkDistributed computingEngineering

Abstract

fetched live from OpenAlex

Neural networks have been widely utilized for wireless communication optimizations. In most of the literature, a dedicated neural network is trained for each specific optimization problem. However, under many scenarios, several distinct objectives are worth optimizing on the same wireless environment. Instead of exhaustively training a new model for every objective, it is more efficient to exploit the correlations between these objectives to train models with shared model parameters and feature representations. In the deep learning literature, transfer learning has been proposed to encourage knowledge transfer among models solving correlated problems. Unlike a majority of transfer learning applications where the high level features are relatively easy to locate in the neural networks, this paper considers wireless communication problems, in which it is much more difficult to identify high level features transferable to correlated tasks. To address this issue, this paper proposes to add an additional reconstruction loss when training the model. This new loss is for reconstructing the problem inputs starting from a selected neural network hidden layer. This approach encourages the features learnt to be general and descriptive about the inputs, instead of being solely responsible for minimizing the specific task-based loss. When a new objective is to be optimized, these features can be readily used for transfer learning. Simulation results in device-to-device wireless network power allocation optimization suggest that the proposed approach is highly efficient in data and model complexity, resilient to over-fitting, and supports competitive optimization performances.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.

Full frame distilled prediction

Teacher imitation

Not calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.

metaresearch head score (Codex)0.001
metaresearch head score (Gemma)0.000
Version: codex-gemma-dda1882f352aValidation status: machine_predicted_unvalidated
Candidate categoriesMeta-epidemiology (narrow), Science and technology studies, Insufficient payload (model declined to judge)
Consensus categoriesnone
DomainCandidate signal: none · Consensus signal: none
Study designCandidate signal: Theoretical or conceptual · Consensus signal: none
GenreCandidate signal: Empirical · Consensus signal: none
Teacher disagreement score0.898
Threshold uncertainty score1.000

Codex and Gemma teacher scores by category

CategoryCodexGemma
Metaresearch0.0010.000
Meta-epidemiology (narrow)0.0000.000
Meta-epidemiology (broad)0.0000.000
Bibliometrics0.0000.002
Science and technology studies0.0020.000
Scholarly communication0.0000.001
Open science0.0040.001
Research integrity0.0000.001
Insufficient payload (model declined to judge)0.0010.000

Machine scores (provisional)

The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.

Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.

Opus teacher head0.031
GPT teacher head0.262
Teacher spread0.231 · how far apart the two teachers sit on this one work
Validation statusscore_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it