Uppmärksamhetsmekanismer för textgenerering

Djupinlärning för text med PyTorch

Shubham Jain

Instructor

Tvetydighet i textbearbetning

"The monkey ate that banana because it was too hungry"
Vad syftar ordet "it" på?

Människa kontra maskin

Uppmärksamhetsmekanismer

Tilldelar ord vikt
Säkerställer att maskinens tolkning stämmer överens med mänsklig förståelse

Uppmärksamhetsdiagram

¹ Xie, Huiqiang & Qin, Zhijin & Li, Geoffrey & Juang, Biing-Hwang. (2020). Deep Learning Enabled Semantic Communication Systems

Self-attention och multi-head attention

Self-Attention: tilldelar ord i en mening betydelse
- "The cat, which was on the roof, was scared"
- Kopplar "was scared" till "The cat"
Multi-Head Attention: som flera strålkastare – fångar olika aspekter
- Förstår att "was scared" kan relatera till
- "The cat", "the roof" eller "was on"

Uppmärksamhetsmekanism – ordförråd och data

data = ["the cat sat on the mat", ...]

vocab = set(' '.join(data).split())

word_to_ix = {word: i for i, word in enumerate(vocab)}
ix_to_word = {i: word for word, i in word_to_ix.items()}

pairs = [sentence.split() for sentence in data]
input_data = [[word_to_ix[word] for word in sentence[:-1]] for sentence in pairs]
target_data = [word_to_ix[sentence[-1]] for sentence in pairs]
inputs = [torch.tensor(seq, dtype=torch.long) for seq in input_data]
targets = torch.tensor(target_data, dtype=torch.long)

Modelldefinition

embedding_dim = 10
hidden_dim = 16


class RNNWithAttentionModel(nn.Module):
    def __init__(self):
        super(RNNWithAttentionModel, self).__init__()

        self.embeddings = nn.Embedding(vocab_size, embedding_dim)
        self.rnn = nn.RNN(embedding_dim, hidden_dim, batch_first=True)

        self.attention = nn.Linear(hidden_dim, 1)

        self.fc = nn.Linear(hidden_dim, vocab_size)

Framåtpropagering med uppmärksamhet

def forward(self, x):
    x = self.embeddings(x)
    out, _ = self.rnn(x)

    attn_weights = torch.nn.functional.softmax(self.attention(out).squeeze(2), 
                                               dim=1)

    context = torch.sum(attn_weights.unsqueeze(2) * out, dim=1)
    out = self.fc(context)
    return out

def pad_sequences(batch): 
    max_len = max([len(seq) for seq in batch])
    return torch.stack([torch.cat([seq, torch.zeros(max_len-len(seq)).long()]) 
                        for seq in batch])

Träningsförberedelse

criterion = nn.CrossEntropyLoss()

attention_model = RNNWithAttentionModel()
optimizer = torch.optim.Adam(attention_model.parameters(), lr=0.01)

for epoch in range(300):
    attention_model.train()
    optimizer.zero_grad()

    padded_inputs = pad_sequences(inputs)
    outputs = attention_model(padded_inputs)

    loss = criterion(outputs, targets)
    loss.backward()
    optimizer.step()

Modellutvärdering

for input_seq, target in zip(input_data, target_data):
    input_test = torch.tensor(input_seq, dtype=torch.long).unsqueeze(0)

    attention_model.eval()
    attention_output = attention_model(input_test)

    attention_prediction = ix_to_word[torch.argmax(attention_output).item()]

    print(f"\nInput: {' '.join([ix_to_word[ix] for ix in input_seq])}")
    print(f"Target: {ix_to_word[target]}")
    print(f"RNN with Attention prediction: {attention_prediction}")

Input: the cat sat on the
Target: mat
RNN with Attention prediction: mat

Nu kör vi en övning!

Djupinlärning för text med PyTorch