Guide-yakabatana mibvunzo · Zvakaoma Nhanho

Sparse Autoencoders yeMibvunzo Yekududzira

Inotarisa sparse autoencoder kuvakazve uye sparsity, activation maficha, hupamhi hwekuwedzera, kupatsanurwa kwezvinhu uye causal kutungamira kuyedza.

Nzira dzekutungamira dzinoenderanaSparse Autoencoders Yekududzira
Mubvunzo 1 ye 8

Chii chinangwa chikuru chekudzidzisa sparse autoencoder pane activation yemuenzaniso?

Tarisa mubvunzo wese8 mibvunzo

Mhinduro uye tsananguro zvinoonekwa paunenge uchitora bvunzo.

  1. Chii chinangwa chikuru chekudzidzisa sparse autoencoder pane activation yemuenzaniso?
    • Kuodza ma actitions akapokana kuita zvinodudzirwa, kazhinji-zvakasarudzika
    • Kumanikidza modhi kuti iite nekukurumidza
    • Kushandura mavara pakati pemitauro
    • Kuti uwedzere kurongeka kweiyo modhi pamabhenji
  2. How does an SAE’s sparsity penalty affect hidden-unit activations?
    • It guarantees each unit represents exactly one concept
    • It encourages only a small subset of the wider hidden layer to activate for an input
    • It makes every hidden unit activate equally
    • It removes the reconstruction objective
  3. Zvingangove zvingani zvinhu zvakatorwa Anthropic pakuyera SAEs kusvika Claude 3 Sonnet muna 2024?
    • Vanenge 500
    • Vanenge 2 000
    • Vanenge mamiriyoni makumi matatu nemana
    • Anenge mabhiriyoni zana
  4. Chii chakaratidza 'Golden Gate Claude'?
    • Kuti maSAE anogona chete kudzidziswa pamifananidzo
    • Kuti iyo modhi haichakwanisa kuburitsa zvinyorwa
    • Izvo zvimiro hazvina simba pane zvinobuda
    • Iko kusungirira chinhu chimwe chete kumusoro kunotungamira maitiro eiyo modhi
  5. Sei iyo yakavanzika layer muSAE inowanzo WIDER pane iyo activation yainovakazve?
    • Nekuti pfungwa dzakaiswa muzvikamu zvishoma pane zvadzinoda chaizvo, saka mamwe mayunitsi anobatsira kuaparadzanisa.
    • Kuita kuti kudzidzisa kudhure
    • Nekuti mapatya akakura anogara akanyatsojeka
    • Kuti uderedze huwandu hwema parameters
  6. Ndeipi inozivikanwa kudzikiswa kweazvino sparse autoencoders?
    • Havagone kudzidziswa ne gradient descent
    • Feature splitting, apo imwe pfungwa inotsemuka kuita akawanda akasiyana maficha
    • Ivo chete vanoshanda pane convolutional network
    • Vanobvisa kudiwa kwechero munhu mazita
  7. Which two objectives are commonly combined when training a sparse autoencoder?
    • A policy gradient and value loss
    • A next-token loss and a contrastive loss
    • Reconstruction error and a penalty on hidden activations
    • A calibration error and classification accuracy
  8. What does a TopK sparse autoencoder do to encourage sparsity?
    • Keeps the K largest input pixels and discards the decoder
    • Sets all encoder weights to zero
    • Trains K different tokenizers
    • Keeps only the K largest hidden activations for an input